agora inbox for [email protected]help / color / mirror / Atom feed
[PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. 178+ messages / 3 participants [nested] [flat]
* [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. @ 2022-06-17 01:33 Andres Freund <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Andres Freund @ 2022-06-17 01:33 UTC (permalink / raw) Author: Reviewed-By: Discussion: https://postgr.es/m/ Backpatch: --- src/include/executor/execExpr.h | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/include/executor/execExpr.h b/src/include/executor/execExpr.h index 2fba1450c65..5bfa7e9bdd4 100644 --- a/src/include/executor/execExpr.h +++ b/src/include/executor/execExpr.h @@ -22,6 +22,8 @@ struct ExprEvalStep; struct SubscriptingRefState; struct ScalarArrayOpExprHashTable; struct JsonbValue; +struct JsonExprState; +struct JsonConstructorExprState; /* Bits in ExprState->flags (see also execnodes.h for public flag bits): */ /* expression's interpreter has been initialized */ -- 2.35.1.677.gabf474a5dd --rftf6si7y32exfgc-- ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v3] Provide relfilenode statistics @ 2023-11-16 02:30 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Bertrand Drouvot @ 2023-11-16 02:30 UTC (permalink / raw) We currently don’t have writes counters for relations. The reason is that we don’t have the relation OID when writing buffers out. Tracking writes per relfilenode would allow us to track/consolidate writes per relation. relfilenode stats is also beneficial for the "Split index and table statistics into different types of stats" work in progress: it would allow us to avoid additional branches in some situations. === Remarks === This is a POC patch. There is still work to do: there is more places we should add relfilenode counters, create more APIS to retrieve the relfilenode stats, the patch takes care of rewrite generated by TRUNCATE but there is more to care about like CLUSTER,VACUUM FULL. The new logic to retrieve stats in pg_statio_all_tables has been implemented only for the new blocks_written stat (we'd need to do the same for the existing buffer read / buffer hit if we agree on the approach implemented here). The goal of this patch is to start the discussion and agree on the design before moving forward. --- src/backend/access/rmgrdesc/xactdesc.c | 5 +- src/backend/catalog/storage.c | 8 ++ src/backend/catalog/system_functions.sql | 2 +- src/backend/catalog/system_views.sql | 5 +- src/backend/postmaster/checkpointer.c | 5 + src/backend/storage/buffer/bufmgr.c | 6 +- src/backend/storage/smgr/md.c | 7 ++ src/backend/utils/activity/pgstat.c | 39 ++++-- src/backend/utils/activity/pgstat_database.c | 12 +- src/backend/utils/activity/pgstat_function.c | 13 +- src/backend/utils/activity/pgstat_relation.c | 112 ++++++++++++++++-- src/backend/utils/activity/pgstat_replslot.c | 13 +- src/backend/utils/activity/pgstat_shmem.c | 19 ++- .../utils/activity/pgstat_subscription.c | 12 +- src/backend/utils/activity/pgstat_xact.c | 60 +++++++--- src/backend/utils/adt/pgstatfuncs.c | 34 +++++- src/include/access/tableam.h | 19 +++ src/include/access/xact.h | 1 + src/include/catalog/pg_proc.dat | 14 ++- src/include/pgstat.h | 37 ++++-- src/include/utils/pgstat_internal.h | 34 ++++-- src/test/recovery/t/029_stats_restart.pl | 40 +++---- .../recovery/t/030_stats_cleanup_replica.pl | 6 +- src/test/regress/expected/rules.out | 12 +- src/test/regress/expected/stats.out | 30 ++--- src/test/regress/sql/stats.sql | 30 ++--- src/test/subscription/t/026_stats.pl | 4 +- src/tools/pgindent/typedefs.list | 1 + 28 files changed, 424 insertions(+), 156 deletions(-) 4.4% src/backend/catalog/ 46.2% src/backend/utils/activity/ 6.3% src/backend/utils/adt/ 3.6% src/backend/ 3.2% src/include/access/ 3.2% src/include/catalog/ 6.0% src/include/utils/ 6.6% src/include/ 11.7% src/test/recovery/t/ 5.3% src/test/regress/expected/ 3.0% src/ diff --git a/src/backend/access/rmgrdesc/xactdesc.c b/src/backend/access/rmgrdesc/xactdesc.c index dccca201e0..c02b079645 100644 --- a/src/backend/access/rmgrdesc/xactdesc.c +++ b/src/backend/access/rmgrdesc/xactdesc.c @@ -319,10 +319,11 @@ xact_desc_stats(StringInfo buf, const char *label, appendStringInfo(buf, "; %sdropped stats:", label); for (i = 0; i < ndropped; i++) { - appendStringInfo(buf, " %d/%u/%u", + appendStringInfo(buf, " %d/%u/%u/%u", dropped_stats[i].kind, dropped_stats[i].dboid, - dropped_stats[i].objoid); + dropped_stats[i].objoid, + dropped_stats[i].relfile); } } } diff --git a/src/backend/catalog/storage.c b/src/backend/catalog/storage.c index f56b3cc0f2..db6107cd90 100644 --- a/src/backend/catalog/storage.c +++ b/src/backend/catalog/storage.c @@ -33,6 +33,7 @@ #include "storage/smgr.h" #include "utils/hsearch.h" #include "utils/memutils.h" +#include "utils/pgstat_internal.h" #include "utils/rel.h" /* GUC variables */ @@ -152,6 +153,7 @@ RelationCreateStorage(RelFileLocator rlocator, char relpersistence, if (needs_wal) log_smgrcreate(&srel->smgr_rlocator.locator, MAIN_FORKNUM); + pgstat_create_transactional(PGSTAT_KIND_RELFILENODE, rlocator.dbOid, rlocator.spcOid, rlocator.relNumber); /* * Add the relation to the list of stuff to delete at abort, if we are * asked to do so. @@ -227,6 +229,8 @@ RelationDropStorage(Relation rel) * for now I'll keep the logic simple. */ + pgstat_drop_transactional(PGSTAT_KIND_RELFILENODE, rel->rd_locator.dbOid, rel->rd_locator.spcOid, rel->rd_locator.relNumber); + RelationCloseSmgr(rel); } @@ -253,6 +257,9 @@ RelationPreserveStorage(RelFileLocator rlocator, bool atCommit) PendingRelDelete *pending; PendingRelDelete *prev; PendingRelDelete *next; + PgStat_SubXactStatus *xact_state; + + xact_state = pgStatXactStack; prev = NULL; for (pending = pendingDeletes; pending != NULL; pending = next) @@ -267,6 +274,7 @@ RelationPreserveStorage(RelFileLocator rlocator, bool atCommit) else pendingDeletes = next; pfree(pending); + PgStat_RemoveRelFileNodeFromDroppedStats(xact_state, rlocator); /* prev does not change */ } else diff --git a/src/backend/catalog/system_functions.sql b/src/backend/catalog/system_functions.sql index 623b9539b1..ec60ef72e3 100644 --- a/src/backend/catalog/system_functions.sql +++ b/src/backend/catalog/system_functions.sql @@ -684,7 +684,7 @@ REVOKE EXECUTE ON FUNCTION pg_stat_reset_single_function_counters(oid) FROM publ REVOKE EXECUTE ON FUNCTION pg_stat_reset_replication_slot(text) FROM public; -REVOKE EXECUTE ON FUNCTION pg_stat_have_stats(text, oid, oid) FROM public; +REVOKE EXECUTE ON FUNCTION pg_stat_have_stats(text, oid, oid, oid) FROM public; REVOKE EXECUTE ON FUNCTION pg_stat_reset_subscription_stats(oid) FROM public; diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql index 19cabc9a47..a7b13d877b 100644 --- a/src/backend/catalog/system_views.sql +++ b/src/backend/catalog/system_views.sql @@ -746,6 +746,7 @@ CREATE VIEW pg_statio_all_tables AS C.relname AS relname, pg_stat_get_blocks_fetched(C.oid) - pg_stat_get_blocks_hit(C.oid) AS heap_blks_read, + pg_stat_get_blocks_written(C.oid) + pg_stat_get_relfilenode_blocks_written(d.oid, CASE WHEN C.reltablespace <> 0 THEN C.reltablespace ELSE d.dattablespace END, C.relfilenode) AS heap_blks_written, pg_stat_get_blocks_hit(C.oid) AS heap_blks_hit, I.idx_blks_read AS idx_blks_read, I.idx_blks_hit AS idx_blks_hit, @@ -754,7 +755,7 @@ CREATE VIEW pg_statio_all_tables AS pg_stat_get_blocks_hit(T.oid) AS toast_blks_hit, X.idx_blks_read AS tidx_blks_read, X.idx_blks_hit AS tidx_blks_hit - FROM pg_class C LEFT JOIN + FROM pg_database d, pg_class C LEFT JOIN pg_class T ON C.reltoastrelid = T.oid LEFT JOIN pg_namespace N ON (N.oid = C.relnamespace) LEFT JOIN LATERAL ( @@ -771,7 +772,7 @@ CREATE VIEW pg_statio_all_tables AS sum(pg_stat_get_blocks_hit(indexrelid))::bigint AS idx_blks_hit FROM pg_index WHERE indrelid = T.oid ) X ON true - WHERE C.relkind IN ('r', 't', 'm'); + WHERE C.relkind IN ('r', 't', 'm') AND d.datname = current_database(); CREATE VIEW pg_statio_sys_tables AS SELECT * FROM pg_statio_all_tables diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c index 199f008bcd..7f13a840c4 100644 --- a/src/backend/postmaster/checkpointer.c +++ b/src/backend/postmaster/checkpointer.c @@ -519,6 +519,11 @@ CheckpointerMain(char *startup_data, size_t startup_data_len) /* Report pending statistics to the cumulative stats system */ pgstat_report_checkpointer(); pgstat_report_wal(true); + /* + * No need to check for transaction state in checkpointer before + * calling pgstat_report_stat(). + */ + pgstat_report_stat(true); /* * If any checkpoint flags have been set, redo the loop to handle the diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c index 4415ba648a..1ac73f672b 100644 --- a/src/backend/storage/buffer/bufmgr.c +++ b/src/backend/storage/buffer/bufmgr.c @@ -1186,9 +1186,9 @@ PinBufferForBlock(Relation rel, * WaitReadBuffers() (so, not for hits, and not for buffers that are * zeroed instead), the per-relation stats always count them. */ - pgstat_count_buffer_read(rel); + pgstat_report_relfilenode_buffer_read(rel); if (*foundPtr) - pgstat_count_buffer_hit(rel); + pgstat_report_relfilenode_buffer_hit(rel); } if (*foundPtr) { @@ -3907,6 +3907,8 @@ FlushBuffer(BufferDesc *buf, SMgrRelation reln, IOObject io_object, pgBufferUsage.shared_blks_written++; + pgstat_report_relfilenode_blks_written(reln->smgr_rlocator.locator); + /* * Mark the buffer as clean (unless BM_JUST_DIRTIED has become set) and * end the BM_IO_IN_PROGRESS state. diff --git a/src/backend/storage/smgr/md.c b/src/backend/storage/smgr/md.c index 6796756358..5bc5fc65cd 100644 --- a/src/backend/storage/smgr/md.c +++ b/src/backend/storage/smgr/md.c @@ -1447,12 +1447,16 @@ DropRelationFiles(RelFileLocator *delrels, int ndelrels, bool isRedo) { SMgrRelation *srels; int i; + int not_freed_count = 0; srels = palloc(sizeof(SMgrRelation) * ndelrels); for (i = 0; i < ndelrels; i++) { SMgrRelation srel = smgropen(delrels[i], INVALID_PROC_NUMBER); + if (!pgstat_drop_entry(PGSTAT_KIND_RELFILENODE, delrels[i].dbOid, delrels[i].spcOid, delrels[i].relNumber)) + not_freed_count++; + if (isRedo) { ForkNumber fork; @@ -1463,6 +1467,9 @@ DropRelationFiles(RelFileLocator *delrels, int ndelrels, bool isRedo) srels[i] = srel; } + if (not_freed_count > 0) + pgstat_request_entry_refs_gc(); + smgrdounlinkall(srels, ndelrels, isRedo); for (i = 0; i < ndelrels; i++) diff --git a/src/backend/utils/activity/pgstat.c b/src/backend/utils/activity/pgstat.c index b2ca3f39b7..035fdb2aa1 100644 --- a/src/backend/utils/activity/pgstat.c +++ b/src/backend/utils/activity/pgstat.c @@ -308,6 +308,19 @@ static const PgStat_KindInfo pgstat_kind_builtin_infos[PGSTAT_KIND_BUILTIN_SIZE] .delete_pending_cb = pgstat_relation_delete_pending_cb, }, + [PGSTAT_KIND_RELFILENODE] = { + .name = "relfilenode", + + .fixed_amount = false, + + .shared_size = sizeof(PgStatShared_RelFileNode), + .shared_data_off = offsetof(PgStatShared_RelFileNode, stats), + .shared_data_len = sizeof(((PgStatShared_RelFileNode *) 0)->stats), + .pending_size = sizeof(PgStat_StatRelFileNodeEntry), + + .flush_pending_cb = pgstat_relfilenode_flush_cb, + }, + [PGSTAT_KIND_FUNCTION] = { .name = "function", @@ -717,7 +730,7 @@ pgstat_report_stat(bool force) partial_flush = false; - /* flush database / relation / function / ... stats */ + /* flush database / relation / function / relfilenode / ... stats */ partial_flush |= pgstat_flush_pending_entries(nowait); /* flush IO stats */ @@ -797,7 +810,7 @@ pgstat_reset_counters(void) * GRANT system. */ void -pgstat_reset(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_reset(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { const PgStat_KindInfo *kind_info = pgstat_get_kind_info(kind); TimestampTz ts = GetCurrentTimestamp(); @@ -806,7 +819,7 @@ pgstat_reset(PgStat_Kind kind, Oid dboid, Oid objoid) Assert(!pgstat_get_kind_info(kind)->fixed_amount); /* reset the "single counter" */ - pgstat_reset_entry(kind, dboid, objoid, ts); + pgstat_reset_entry(kind, dboid, objoid, relfile, ts); if (!kind_info->accessed_across_databases) pgstat_reset_database_timestamp(dboid, ts); @@ -877,7 +890,7 @@ pgstat_clear_snapshot(void) } void * -pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { PgStat_HashKey key; PgStat_EntryRef *entry_ref; @@ -893,6 +906,7 @@ pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid) key.kind = kind; key.dboid = dboid; key.objoid = objoid; + key.relfile = relfile; /* if we need to build a full snapshot, do so */ if (pgstat_fetch_consistency == PGSTAT_FETCH_CONSISTENCY_SNAPSHOT) @@ -918,7 +932,7 @@ pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid) pgStatLocal.snapshot.mode = pgstat_fetch_consistency; - entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, false, NULL); + entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, relfile, false, NULL); if (entry_ref == NULL || entry_ref->shared_entry->dropped) { @@ -987,13 +1001,13 @@ pgstat_get_stat_snapshot_timestamp(bool *have_snapshot) } bool -pgstat_have_entry(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_have_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { /* fixed-numbered stats always exist */ if (pgstat_get_kind_info(kind)->fixed_amount) return true; - return pgstat_get_entry_ref(kind, dboid, objoid, false, NULL) != NULL; + return pgstat_get_entry_ref(kind, dboid, objoid, relfile, false, NULL) != NULL; } /* @@ -1208,7 +1222,8 @@ pgstat_build_snapshot_fixed(PgStat_Kind kind) * created, false otherwise. */ PgStat_EntryRef * -pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, bool *created_entry) +pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, + RelFileNumber relfile, bool *created_entry) { PgStat_EntryRef *entry_ref; @@ -1223,7 +1238,7 @@ pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, bool *created ALLOCSET_SMALL_SIZES); } - entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, + entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, relfile, true, created_entry); if (entry_ref->pending == NULL) @@ -1246,11 +1261,11 @@ pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, bool *created * that it shouldn't be needed. */ PgStat_EntryRef * -pgstat_fetch_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_fetch_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { PgStat_EntryRef *entry_ref; - entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, false, NULL); + entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, relfile, false, NULL); if (entry_ref == NULL || entry_ref->pending == NULL) return NULL; @@ -1279,7 +1294,7 @@ pgstat_delete_pending_entry(PgStat_EntryRef *entry_ref) } /* - * Flush out pending stats for database objects (databases, relations, + * Flush out pending stats for database objects (databases, relations, relfilenodes, * functions). */ static bool diff --git a/src/backend/utils/activity/pgstat_database.c b/src/backend/utils/activity/pgstat_database.c index 29bc090974..cf77f2dbdb 100644 --- a/src/backend/utils/activity/pgstat_database.c +++ b/src/backend/utils/activity/pgstat_database.c @@ -43,7 +43,7 @@ static PgStat_Counter pgLastSessionReportTime = 0; void pgstat_drop_database(Oid databaseid) { - pgstat_drop_transactional(PGSTAT_KIND_DATABASE, databaseid, InvalidOid); + pgstat_drop_transactional(PGSTAT_KIND_DATABASE, databaseid, InvalidOid, InvalidOid); } /* @@ -66,7 +66,7 @@ pgstat_report_autovac(Oid dboid) * operation so it doesn't matter if we get blocked here a little. */ entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, - dboid, InvalidOid, false); + dboid, InvalidOid, InvalidOid, false); dbentry = (PgStatShared_Database *) entry_ref->shared_stats; dbentry->stats.last_autovac_time = GetCurrentTimestamp(); @@ -150,7 +150,7 @@ pgstat_report_checksum_failures_in_db(Oid dboid, int failurecount) * common enough for that to be a problem. */ entry_ref = - pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, dboid, InvalidOid, false); + pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, dboid, InvalidOid, InvalidOid, false); sharedent = (PgStatShared_Database *) entry_ref->shared_stats; sharedent->stats.checksum_failures += failurecount; @@ -242,7 +242,7 @@ PgStat_StatDBEntry * pgstat_fetch_stat_dbentry(Oid dboid) { return (PgStat_StatDBEntry *) - pgstat_fetch_entry(PGSTAT_KIND_DATABASE, dboid, InvalidOid); + pgstat_fetch_entry(PGSTAT_KIND_DATABASE, dboid, InvalidOid, InvalidOid); } void @@ -341,7 +341,7 @@ pgstat_prep_database_pending(Oid dboid) Assert(!OidIsValid(dboid) || OidIsValid(MyDatabaseId)); entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_DATABASE, dboid, InvalidOid, - NULL); + InvalidOid, NULL); return entry_ref->pending; } @@ -357,7 +357,7 @@ pgstat_reset_database_timestamp(Oid dboid, TimestampTz ts) PgStatShared_Database *dbentry; dbref = pgstat_get_entry_ref_locked(PGSTAT_KIND_DATABASE, MyDatabaseId, InvalidOid, - false); + InvalidOid, false); dbentry = (PgStatShared_Database *) dbref->shared_stats; dbentry->stats.stat_reset_timestamp = ts; diff --git a/src/backend/utils/activity/pgstat_function.c b/src/backend/utils/activity/pgstat_function.c index d26da551a4..440e44e300 100644 --- a/src/backend/utils/activity/pgstat_function.c +++ b/src/backend/utils/activity/pgstat_function.c @@ -46,7 +46,8 @@ pgstat_create_function(Oid proid) { pgstat_create_transactional(PGSTAT_KIND_FUNCTION, MyDatabaseId, - proid); + proid, + InvalidOid); } /* @@ -61,7 +62,8 @@ pgstat_drop_function(Oid proid) { pgstat_drop_transactional(PGSTAT_KIND_FUNCTION, MyDatabaseId, - proid); + proid, + InvalidOid); } /* @@ -86,6 +88,7 @@ pgstat_init_function_usage(FunctionCallInfo fcinfo, entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, fcinfo->flinfo->fn_oid, + InvalidOid, &created_entry); /* @@ -113,7 +116,7 @@ pgstat_init_function_usage(FunctionCallInfo fcinfo, if (!SearchSysCacheExists1(PROCOID, ObjectIdGetDatum(fcinfo->flinfo->fn_oid))) { pgstat_drop_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, - fcinfo->flinfo->fn_oid); + fcinfo->flinfo->fn_oid, InvalidOid); ereport(ERROR, errcode(ERRCODE_UNDEFINED_FUNCTION), errmsg("function call to dropped function")); } @@ -224,7 +227,7 @@ find_funcstat_entry(Oid func_id) { PgStat_EntryRef *entry_ref; - entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, func_id); + entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, func_id, InvalidOid); if (entry_ref) return entry_ref->pending; @@ -239,5 +242,5 @@ PgStat_StatFuncEntry * pgstat_fetch_stat_funcentry(Oid func_id) { return (PgStat_StatFuncEntry *) - pgstat_fetch_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, func_id); + pgstat_fetch_entry(PGSTAT_KIND_FUNCTION, MyDatabaseId, func_id, InvalidOid); } diff --git a/src/backend/utils/activity/pgstat_relation.c b/src/backend/utils/activity/pgstat_relation.c index 8a3f7d434c..136dd6c85b 100644 --- a/src/backend/utils/activity/pgstat_relation.c +++ b/src/backend/utils/activity/pgstat_relation.c @@ -44,6 +44,7 @@ typedef struct TwoPhasePgStatRecord static PgStat_TableStatus *pgstat_prep_relation_pending(Oid rel_id, bool isshared); +PgStat_StatRelFileNodeEntry *pgstat_prep_relfilenode_pending(RelFileLocator locator); static void add_tabstat_xact_level(PgStat_TableStatus *pgstat_info, int nest_level); static void ensure_tabstat_xact_level(PgStat_TableStatus *pgstat_info); static void save_truncdrop_counters(PgStat_TableXactStatus *trans, bool is_drop); @@ -69,6 +70,7 @@ pgstat_copy_relation_stats(Relation dst, Relation src) dst_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_RELATION, dst->rd_rel->relisshared ? InvalidOid : MyDatabaseId, RelationGetRelid(dst), + InvalidOid, false); dstshstats = (PgStatShared_Relation *) dst_ref->shared_stats; @@ -170,7 +172,7 @@ pgstat_create_relation(Relation rel) { pgstat_create_transactional(PGSTAT_KIND_RELATION, rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId, - RelationGetRelid(rel)); + RelationGetRelid(rel), InvalidOid); } /* @@ -184,7 +186,7 @@ pgstat_drop_relation(Relation rel) pgstat_drop_transactional(PGSTAT_KIND_RELATION, rel->rd_rel->relisshared ? InvalidOid : MyDatabaseId, - RelationGetRelid(rel)); + RelationGetRelid(rel), InvalidOid); if (!pgstat_should_count_relation(rel)) return; @@ -225,7 +227,7 @@ pgstat_report_vacuum(Oid tableoid, bool shared, /* block acquiring lock for the same reason as pgstat_report_autovac() */ entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_RELATION, - dboid, tableoid, false); + dboid, tableoid, InvalidOid, false); shtabentry = (PgStatShared_Relation *) entry_ref->shared_stats; tabentry = &shtabentry->stats; @@ -318,6 +320,7 @@ pgstat_report_analyze(Relation rel, /* block acquiring lock for the same reason as pgstat_report_autovac() */ entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_RELATION, dboid, RelationGetRelid(rel), + InvalidOid, false); /* can't get dropped while accessed */ Assert(entry_ref != NULL && entry_ref->shared_stats != NULL); @@ -458,6 +461,19 @@ pgstat_fetch_stat_tabentry(Oid relid) return pgstat_fetch_stat_tabentry_ext(IsSharedRelation(relid), relid); } +/* + * Support function for the SQL-callable pgstat* functions. Returns + * the collected statistics for one relfilenode or NULL. NULL doesn't mean + * that the relfilenode doesn't exist, just that there are no statistics, so the + * caller is better off to report ZERO instead. + */ +PgStat_StatRelFileNodeEntry * +pgstat_fetch_stat_relfilenodeentry(Oid dboid, Oid spcOid, RelFileNumber relfile) +{ + return (PgStat_StatRelFileNodeEntry *) + pgstat_fetch_entry(PGSTAT_KIND_RELFILENODE, dboid, spcOid, relfile); +} + /* * More efficient version of pgstat_fetch_stat_tabentry(), allowing to specify * whether the to-be-accessed table is a shared relation or not. @@ -468,7 +484,7 @@ pgstat_fetch_stat_tabentry_ext(bool shared, Oid reloid) Oid dboid = (shared ? InvalidOid : MyDatabaseId); return (PgStat_StatTabEntry *) - pgstat_fetch_entry(PGSTAT_KIND_RELATION, dboid, reloid); + pgstat_fetch_entry(PGSTAT_KIND_RELATION, dboid, reloid, InvalidOid); } /* @@ -491,10 +507,10 @@ find_tabstat_entry(Oid rel_id) PgStat_TableStatus *tabentry = NULL; PgStat_TableStatus *tablestatus = NULL; - entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_RELATION, MyDatabaseId, rel_id); + entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_RELATION, MyDatabaseId, rel_id, InvalidOid); if (!entry_ref) { - entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_RELATION, InvalidOid, rel_id); + entry_ref = pgstat_fetch_pending_entry(PGSTAT_KIND_RELATION, InvalidOid, rel_id, InvalidOid); if (!entry_ref) return tablestatus; } @@ -881,6 +897,38 @@ pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) return true; } +/* + * Flush out pending stats for the relfilenode entry + * + * If nowait is true, this function returns false if lock could not + * immediately acquired, otherwise true is returned. + */ +bool +pgstat_relfilenode_flush_cb(PgStat_EntryRef *entry_ref, bool nowait) +{ + PgStatShared_RelFileNode *sharedent; + PgStat_StatRelFileNodeEntry *pendingent; + + pendingent = (PgStat_StatRelFileNodeEntry *) entry_ref->pending; + sharedent = (PgStatShared_RelFileNode *) entry_ref->shared_stats; + + if (!pgstat_lock_entry(entry_ref, nowait)) + return false; + +#define PGSTAT_ACCUM_RELFILENODECOUNT(item) \ + (sharedent)->stats.item += (pendingent)->item + + PGSTAT_ACCUM_RELFILENODECOUNT(blocks_fetched); + PGSTAT_ACCUM_RELFILENODECOUNT(blocks_hit); + PGSTAT_ACCUM_RELFILENODECOUNT(blocks_written); + + pgstat_unlock_entry(entry_ref); + + memset(pendingent, 0, sizeof(*pendingent)); + + return true; +} + void pgstat_relation_delete_pending_cb(PgStat_EntryRef *entry_ref) { @@ -902,7 +950,7 @@ pgstat_prep_relation_pending(Oid rel_id, bool isshared) entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_RELATION, isshared ? InvalidOid : MyDatabaseId, - rel_id, NULL); + rel_id, InvalidOid, NULL); pending = entry_ref->pending; pending->id = rel_id; pending->shared = isshared; @@ -910,6 +958,56 @@ pgstat_prep_relation_pending(Oid rel_id, bool isshared) return pending; } +PgStat_StatRelFileNodeEntry * +pgstat_prep_relfilenode_pending(RelFileLocator locator) +{ + PgStat_EntryRef *entry_ref; + + entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_RELFILENODE, locator.dbOid, + locator.spcOid, locator.relNumber, NULL); + + return entry_ref->pending; +} + +void +pgstat_report_relfilenode_blks_written(RelFileLocator locator) +{ + PgStat_StatRelFileNodeEntry *relfileentry = NULL; + + relfileentry = pgstat_prep_relfilenode_pending(locator); + + if (relfileentry) + relfileentry->blocks_written++; +} + +void +pgstat_report_relfilenode_buffer_read(Relation reln) +{ + PgStat_StatRelFileNodeEntry *relfileentry = NULL; + + /* For relation stats to survive after a rewrite */ + pgstat_count_buffer_read(reln); + + relfileentry = pgstat_prep_relfilenode_pending(reln->rd_locator); + + if (relfileentry) + relfileentry->blocks_fetched++; +} + +void +pgstat_report_relfilenode_buffer_hit(Relation reln) +{ + PgStat_StatRelFileNodeEntry *relfileentry = NULL; + + /* For relation stats to survive after a rewrite */ + pgstat_count_buffer_hit(reln); + + relfileentry = pgstat_prep_relfilenode_pending(reln->rd_locator); + + if (relfileentry) + relfileentry->blocks_hit++; +} + /* * add a new (sub)transaction state record */ diff --git a/src/backend/utils/activity/pgstat_replslot.c b/src/backend/utils/activity/pgstat_replslot.c index da11b86744..2e68ed4a09 100644 --- a/src/backend/utils/activity/pgstat_replslot.c +++ b/src/backend/utils/activity/pgstat_replslot.c @@ -62,7 +62,7 @@ pgstat_reset_replslot(const char *name) */ if (SlotIsLogical(slot)) pgstat_reset(PGSTAT_KIND_REPLSLOT, InvalidOid, - ReplicationSlotIndex(slot)); + ReplicationSlotIndex(slot), InvalidOid); LWLockRelease(ReplicationSlotControlLock); } @@ -82,7 +82,7 @@ pgstat_report_replslot(ReplicationSlot *slot, const PgStat_StatReplSlotEntry *re PgStat_StatReplSlotEntry *statent; entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_REPLSLOT, InvalidOid, - ReplicationSlotIndex(slot), false); + ReplicationSlotIndex(slot), InvalidOid, false); shstatent = (PgStatShared_ReplSlot *) entry_ref->shared_stats; statent = &shstatent->stats; @@ -116,7 +116,7 @@ pgstat_create_replslot(ReplicationSlot *slot) Assert(LWLockHeldByMeInMode(ReplicationSlotAllocationLock, LW_EXCLUSIVE)); entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_REPLSLOT, InvalidOid, - ReplicationSlotIndex(slot), false); + ReplicationSlotIndex(slot), InvalidOid, false); shstatent = (PgStatShared_ReplSlot *) entry_ref->shared_stats; /* @@ -146,7 +146,7 @@ void pgstat_acquire_replslot(ReplicationSlot *slot) { pgstat_get_entry_ref(PGSTAT_KIND_REPLSLOT, InvalidOid, - ReplicationSlotIndex(slot), true, NULL); + ReplicationSlotIndex(slot), InvalidOid, true, NULL); } /* @@ -158,7 +158,7 @@ pgstat_drop_replslot(ReplicationSlot *slot) Assert(LWLockHeldByMeInMode(ReplicationSlotAllocationLock, LW_EXCLUSIVE)); if (!pgstat_drop_entry(PGSTAT_KIND_REPLSLOT, InvalidOid, - ReplicationSlotIndex(slot))) + ReplicationSlotIndex(slot), InvalidOid)) pgstat_request_entry_refs_gc(); } @@ -178,7 +178,7 @@ pgstat_fetch_replslot(NameData slotname) if (idx != -1) slotentry = (PgStat_StatReplSlotEntry *) pgstat_fetch_entry(PGSTAT_KIND_REPLSLOT, - InvalidOid, idx); + InvalidOid, idx, InvalidOid); LWLockRelease(ReplicationSlotControlLock); @@ -210,6 +210,7 @@ pgstat_replslot_from_serialized_name_cb(const NameData *name, PgStat_HashKey *ke key->kind = PGSTAT_KIND_REPLSLOT; key->dboid = InvalidOid; key->objoid = idx; + key->relfile = InvalidOid; return true; } diff --git a/src/backend/utils/activity/pgstat_shmem.c b/src/backend/utils/activity/pgstat_shmem.c index fd09b9d988..4890f8b807 100644 --- a/src/backend/utils/activity/pgstat_shmem.c +++ b/src/backend/utils/activity/pgstat_shmem.c @@ -429,10 +429,10 @@ pgstat_get_entry_ref_cached(PgStat_HashKey key, PgStat_EntryRef **entry_ref_p) * if the entry is newly created, false otherwise. */ PgStat_EntryRef * -pgstat_get_entry_ref(PgStat_Kind kind, Oid dboid, Oid objoid, bool create, - bool *created_entry) +pgstat_get_entry_ref(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile, + bool create, bool *created_entry) { - PgStat_HashKey key = {.kind = kind,.dboid = dboid,.objoid = objoid}; + PgStat_HashKey key = {.kind = kind,.dboid = dboid,.objoid = objoid,.relfile = relfile}; PgStatShared_HashEntry *shhashent; PgStatShared_Common *shheader = NULL; PgStat_EntryRef *entry_ref; @@ -645,12 +645,12 @@ pgstat_unlock_entry(PgStat_EntryRef *entry_ref) */ PgStat_EntryRef * pgstat_get_entry_ref_locked(PgStat_Kind kind, Oid dboid, Oid objoid, - bool nowait) + RelFileNumber relfile, bool nowait) { PgStat_EntryRef *entry_ref; /* find shared table stats entry corresponding to the local entry */ - entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, true, NULL); + entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, relfile, true, NULL); /* lock the shared entry to protect the content, skip if failed */ if (!pgstat_lock_entry(entry_ref, nowait)) @@ -905,9 +905,9 @@ pgstat_drop_database_and_contents(Oid dboid) * pgstat_gc_entry_refs(). */ bool -pgstat_drop_entry(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_drop_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { - PgStat_HashKey key = {.kind = kind,.dboid = dboid,.objoid = objoid}; + PgStat_HashKey key = {.kind = kind,.dboid = dboid,.objoid = objoid,.relfile = relfile}; PgStatShared_HashEntry *shent; bool freed = true; @@ -980,13 +980,12 @@ shared_stat_reset_contents(PgStat_Kind kind, PgStatShared_Common *header, * Reset one variable-numbered stats entry. */ void -pgstat_reset_entry(PgStat_Kind kind, Oid dboid, Oid objoid, TimestampTz ts) +pgstat_reset_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile, TimestampTz ts) { PgStat_EntryRef *entry_ref; Assert(!pgstat_get_kind_info(kind)->fixed_amount); - - entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, false, NULL); + entry_ref = pgstat_get_entry_ref(kind, dboid, objoid, relfile, false, NULL); if (!entry_ref || entry_ref->shared_entry->dropped) return; diff --git a/src/backend/utils/activity/pgstat_subscription.c b/src/backend/utils/activity/pgstat_subscription.c index d9af8de658..9b9ab2861b 100644 --- a/src/backend/utils/activity/pgstat_subscription.c +++ b/src/backend/utils/activity/pgstat_subscription.c @@ -30,7 +30,7 @@ pgstat_report_subscription_error(Oid subid, bool is_apply_error) PgStat_BackendSubEntry *pending; entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_SUBSCRIPTION, - InvalidOid, subid, NULL); + InvalidOid, subid, InvalidOid, NULL); pending = entry_ref->pending; if (is_apply_error) @@ -47,12 +47,12 @@ pgstat_create_subscription(Oid subid) { /* Ensures that stats are dropped if transaction rolls back */ pgstat_create_transactional(PGSTAT_KIND_SUBSCRIPTION, - InvalidOid, subid); + InvalidOid, subid, InvalidOid); /* Create and initialize the subscription stats entry */ - pgstat_get_entry_ref(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, + pgstat_get_entry_ref(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, InvalidOid, true, NULL); - pgstat_reset_entry(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, 0); + pgstat_reset_entry(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, InvalidOid, 0); } /* @@ -64,7 +64,7 @@ void pgstat_drop_subscription(Oid subid) { pgstat_drop_transactional(PGSTAT_KIND_SUBSCRIPTION, - InvalidOid, subid); + InvalidOid, subid, InvalidOid); } /* @@ -75,7 +75,7 @@ PgStat_StatSubEntry * pgstat_fetch_stat_subscription(Oid subid) { return (PgStat_StatSubEntry *) - pgstat_fetch_entry(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid); + pgstat_fetch_entry(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, InvalidOid); } /* diff --git a/src/backend/utils/activity/pgstat_xact.c b/src/backend/utils/activity/pgstat_xact.c index 1877d22f14..b25df5112b 100644 --- a/src/backend/utils/activity/pgstat_xact.c +++ b/src/backend/utils/activity/pgstat_xact.c @@ -30,7 +30,7 @@ static void AtEOXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, bool static void AtEOSubXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, bool isCommit, int nestDepth); -static PgStat_SubXactStatus *pgStatXactStack = NULL; +PgStat_SubXactStatus *pgStatXactStack = NULL; /* @@ -84,7 +84,7 @@ AtEOXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, bool isCommit) * Transaction that dropped an object committed. Drop the stats * too. */ - if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid)) + if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid, it->relfile)) not_freed_count++; } else if (!isCommit && pending->is_create) @@ -93,7 +93,7 @@ AtEOXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, bool isCommit) * Transaction that created an object aborted. Drop the stats * associated with the object. */ - if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid)) + if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid, it->relfile)) not_freed_count++; } @@ -105,6 +105,33 @@ AtEOXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, bool isCommit) pgstat_request_entry_refs_gc(); } +/* + * Remove a relfilenode stat from the list of stats to be dropped. + */ +void +PgStat_RemoveRelFileNodeFromDroppedStats(PgStat_SubXactStatus *xact_state, RelFileLocator rlocator) +{ + dlist_mutable_iter iter; + + if (dclist_count(&xact_state->pending_drops) == 0) + return; + + dclist_foreach_modify(iter, &xact_state->pending_drops) + { + PgStat_PendingDroppedStatsItem *pending = + dclist_container(PgStat_PendingDroppedStatsItem, node, iter.cur); + xl_xact_stats_item *it = &pending->item; + + if (it->kind == PGSTAT_KIND_RELFILENODE && it->dboid == rlocator.dbOid + && it->objoid == rlocator.spcOid && it->relfile == rlocator.relNumber) + { + dclist_delete_from(&xact_state->pending_drops, &pending->node); + pfree(pending); + return; + } + } +} + /* * Called from access/transam/xact.c at subtransaction commit/abort. */ @@ -158,7 +185,7 @@ AtEOSubXact_PgStat_DroppedStats(PgStat_SubXactStatus *xact_state, * Subtransaction creating a new stats object aborted. Drop the * stats object. */ - if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid)) + if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid, it->relfile)) not_freed_count++; pfree(pending); } @@ -320,7 +347,11 @@ pgstat_execute_transactional_drops(int ndrops, struct xl_xact_stats_item *items, { xl_xact_stats_item *it = &items[i]; - if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid)) + /* leave it to pgstat_drop_transactional() in RelationDropStorage() */ + if (it->kind == PGSTAT_KIND_RELFILENODE) + continue; + + if (!pgstat_drop_entry(it->kind, it->dboid, it->objoid, it->relfile)) not_freed_count++; } @@ -329,7 +360,7 @@ pgstat_execute_transactional_drops(int ndrops, struct xl_xact_stats_item *items, } static void -create_drop_transactional_internal(PgStat_Kind kind, Oid dboid, Oid objoid, bool is_create) +create_drop_transactional_internal(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile, bool is_create) { int nest_level = GetCurrentTransactionNestLevel(); PgStat_SubXactStatus *xact_state; @@ -342,6 +373,7 @@ create_drop_transactional_internal(PgStat_Kind kind, Oid dboid, Oid objoid, bool drop->item.kind = kind; drop->item.dboid = dboid; drop->item.objoid = objoid; + drop->item.relfile = relfile; dclist_push_tail(&xact_state->pending_drops, &drop->node); } @@ -354,18 +386,18 @@ create_drop_transactional_internal(PgStat_Kind kind, Oid dboid, Oid objoid, bool * dropped. */ void -pgstat_create_transactional(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_create_transactional(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { - if (pgstat_get_entry_ref(kind, dboid, objoid, false, NULL)) + if (pgstat_get_entry_ref(kind, dboid, objoid, relfile, false, NULL)) { ereport(WARNING, - errmsg("resetting existing statistics for kind %s, db=%u, oid=%u", - (pgstat_get_kind_info(kind))->name, dboid, objoid)); + errmsg("resetting existing statistics for kind %s, db=%u, oid=%u, relfile=%u", + (pgstat_get_kind_info(kind))->name, dboid, objoid, relfile)); - pgstat_reset(kind, dboid, objoid); + pgstat_reset(kind, dboid, objoid, relfile); } - create_drop_transactional_internal(kind, dboid, objoid, /* create */ true); + create_drop_transactional_internal(kind, dboid, objoid, relfile, /* create */ true); } /* @@ -376,7 +408,7 @@ pgstat_create_transactional(PgStat_Kind kind, Oid dboid, Oid objoid) * alive. */ void -pgstat_drop_transactional(PgStat_Kind kind, Oid dboid, Oid objoid) +pgstat_drop_transactional(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile) { - create_drop_transactional_internal(kind, dboid, objoid, /* create */ false); + create_drop_transactional_internal(kind, dboid, objoid, relfile, /* create */ false); } diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c index 3221137123..c1d62873a3 100644 --- a/src/backend/utils/adt/pgstatfuncs.c +++ b/src/backend/utils/adt/pgstatfuncs.c @@ -106,6 +106,30 @@ PG_STAT_GET_RELENTRY_INT64(tuples_updated) /* pg_stat_get_vacuum_count */ PG_STAT_GET_RELENTRY_INT64(vacuum_count) +#define PG_STAT_GET_RELFILEENTRY_INT64(stat) \ +Datum \ +CppConcat(pg_stat_get_relfilenode_,stat)(PG_FUNCTION_ARGS) \ +{ \ + Oid dboid = PG_GETARG_OID(0); \ + Oid spcOid = PG_GETARG_OID(1); \ + RelFileNumber relfile = PG_GETARG_OID(2); \ + int64 result; \ + PgStat_StatRelFileNodeEntry *relfileentry; \ + \ + if ((relfileentry = pgstat_fetch_stat_relfilenodeentry(dboid, spcOid, relfile)) == NULL) \ + result = 0; \ + else \ + result = (int64) (relfileentry->stat); \ + \ + PG_RETURN_INT64(result); \ +} + +/* pg_stat_get_relfilenode_blocks_written */ +PG_STAT_GET_RELFILEENTRY_INT64(blocks_written) + +/* pg_stat_get_blocks_written */ +PG_STAT_GET_RELENTRY_INT64(blocks_written) + #define PG_STAT_GET_RELENTRY_TIMESTAMPTZ(stat) \ Datum \ CppConcat(pg_stat_get_,stat)(PG_FUNCTION_ARGS) \ @@ -1752,7 +1776,7 @@ pg_stat_reset_single_table_counters(PG_FUNCTION_ARGS) Oid taboid = PG_GETARG_OID(0); Oid dboid = (IsSharedRelation(taboid) ? InvalidOid : MyDatabaseId); - pgstat_reset(PGSTAT_KIND_RELATION, dboid, taboid); + pgstat_reset(PGSTAT_KIND_RELATION, dboid, taboid, InvalidOid); PG_RETURN_VOID(); } @@ -1762,7 +1786,7 @@ pg_stat_reset_single_function_counters(PG_FUNCTION_ARGS) { Oid funcoid = PG_GETARG_OID(0); - pgstat_reset(PGSTAT_KIND_FUNCTION, MyDatabaseId, funcoid); + pgstat_reset(PGSTAT_KIND_FUNCTION, MyDatabaseId, funcoid, InvalidOid); PG_RETURN_VOID(); } @@ -1820,7 +1844,7 @@ pg_stat_reset_subscription_stats(PG_FUNCTION_ARGS) ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE), errmsg("invalid subscription OID %u", subid))); - pgstat_reset(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid); + pgstat_reset(PGSTAT_KIND_SUBSCRIPTION, InvalidOid, subid, InvalidOid); } PG_RETURN_VOID(); @@ -2028,7 +2052,9 @@ pg_stat_have_stats(PG_FUNCTION_ARGS) char *stats_type = text_to_cstring(PG_GETARG_TEXT_P(0)); Oid dboid = PG_GETARG_OID(1); Oid objoid = PG_GETARG_OID(2); + Oid relfile = PG_GETARG_OID(3); + PgStat_Kind kind = pgstat_get_kind_from_str(stats_type); - PG_RETURN_BOOL(pgstat_have_entry(kind, dboid, objoid)); + PG_RETURN_BOOL(pgstat_have_entry(kind, dboid, objoid, relfile)); } diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h index da661289c1..3614bae63c 100644 --- a/src/include/access/tableam.h +++ b/src/include/access/tableam.h @@ -21,7 +21,9 @@ #include "access/sdir.h" #include "access/xact.h" #include "executor/tuptable.h" +#include "pgstat.h" #include "storage/read_stream.h" +#include "utils/pgstat_internal.h" #include "utils/rel.h" #include "utils/snapshot.h" @@ -1624,6 +1626,23 @@ table_relation_set_new_filelocator(Relation rel, TransactionId *freezeXid, MultiXactId *minmulti) { + PgStat_StatRelFileNodeEntry *relfileentry; + PgStat_StatTabEntry *tabentry = NULL; + PgStat_EntryRef *entry_ref = NULL; + PgStatShared_Relation *shtabentry; + + entry_ref = pgstat_get_entry_ref(PGSTAT_KIND_RELATION, MyDatabaseId, rel->rd_id, InvalidOid, false, NULL); + if (entry_ref) + { + shtabentry = (PgStatShared_Relation *) entry_ref->shared_stats; + tabentry = &shtabentry->stats; + } + + relfileentry = pgstat_fetch_stat_relfilenodeentry(rel->rd_locator.dbOid, rel->rd_locator.spcOid, rel->rd_locator.relNumber); + + if (tabentry && relfileentry) + tabentry->blocks_written += relfileentry->blocks_written; + rel->rd_tableam->relation_set_new_filelocator(rel, newrlocator, persistence, freezeXid, minmulti); diff --git a/src/include/access/xact.h b/src/include/access/xact.h index 6d4439f052..3b9ed65ff6 100644 --- a/src/include/access/xact.h +++ b/src/include/access/xact.h @@ -284,6 +284,7 @@ typedef struct xl_xact_stats_item int kind; Oid dboid; Oid objoid; + RelFileNumber relfile; } xl_xact_stats_item; typedef struct xl_xact_stats_items diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index d36f6001bb..fae89b1e08 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -5407,6 +5407,14 @@ proname => 'pg_stat_get_tuples_updated', provolatile => 's', proparallel => 'r', prorettype => 'int8', proargtypes => 'oid', prosrc => 'pg_stat_get_tuples_updated' }, +{ oid => '9280', descr => 'statistics: number of blocks written', + proname => 'pg_stat_get_relfilenode_blocks_written', provolatile => 's', + proparallel => 'r', + proargtypes => 'oid oid oid', + prorettype => 'int8', + proallargtypes => '{oid,oid,oid,int8}', + proargmodes => '{i,i,i,o}', + prosrc => 'pg_stat_get_relfilenode_blocks_written' }, { oid => '1933', descr => 'statistics: number of tuples deleted', proname => 'pg_stat_get_tuples_deleted', provolatile => 's', proparallel => 'r', prorettype => 'int8', proargtypes => 'oid', @@ -5446,6 +5454,10 @@ proname => 'pg_stat_get_blocks_hit', provolatile => 's', proparallel => 'r', prorettype => 'int8', proargtypes => 'oid', prosrc => 'pg_stat_get_blocks_hit' }, +{ oid => '8438', descr => 'statistics: number of blocks written', + proname => 'pg_stat_get_blocks_written', provolatile => 's', proparallel => 'r', + prorettype => 'int8', proargtypes => 'oid', + prosrc => 'pg_stat_get_blocks_written' }, { oid => '2781', descr => 'statistics: last manual vacuum time for a table', proname => 'pg_stat_get_last_vacuum_time', provolatile => 's', proparallel => 'r', prorettype => 'timestamptz', proargtypes => 'oid', @@ -5532,7 +5544,7 @@ { oid => '6230', descr => 'statistics: check if a stats object exists', proname => 'pg_stat_have_stats', provolatile => 'v', proparallel => 'r', - prorettype => 'bool', proargtypes => 'text oid oid', + prorettype => 'bool', proargtypes => 'text oid oid oid', prosrc => 'pg_stat_have_stats' }, { oid => '6231', descr => 'statistics: information about subscription stats', diff --git a/src/include/pgstat.h b/src/include/pgstat.h index f63159c55c..cfed4f07f5 100644 --- a/src/include/pgstat.h +++ b/src/include/pgstat.h @@ -15,6 +15,7 @@ #include "datatype/timestamp.h" #include "portability/instr_time.h" #include "postmaster/pgarch.h" /* for MAX_XFN_CHARS */ +#include "storage/relfilelocator.h" #include "utils/backend_progress.h" /* for backward compatibility */ #include "utils/backend_status.h" /* for backward compatibility */ #include "utils/relcache.h" @@ -45,17 +46,18 @@ /* stats for variable-numbered objects */ #define PGSTAT_KIND_DATABASE 1 /* database-wide statistics */ #define PGSTAT_KIND_RELATION 2 /* per-table statistics */ -#define PGSTAT_KIND_FUNCTION 3 /* per-function statistics */ -#define PGSTAT_KIND_REPLSLOT 4 /* per-slot statistics */ -#define PGSTAT_KIND_SUBSCRIPTION 5 /* per-subscription statistics */ +#define PGSTAT_KIND_RELFILENODE 3 /* per-relfilenode statistics */ +#define PGSTAT_KIND_FUNCTION 4 /* per-function statistics */ +#define PGSTAT_KIND_REPLSLOT 5 /* per-slot statistics */ +#define PGSTAT_KIND_SUBSCRIPTION 6 /* per-subscription statistics */ /* stats for fixed-numbered objects */ -#define PGSTAT_KIND_ARCHIVER 6 -#define PGSTAT_KIND_BGWRITER 7 -#define PGSTAT_KIND_CHECKPOINTER 8 -#define PGSTAT_KIND_IO 9 -#define PGSTAT_KIND_SLRU 10 -#define PGSTAT_KIND_WAL 11 +#define PGSTAT_KIND_ARCHIVER 7 +#define PGSTAT_KIND_BGWRITER 8 +#define PGSTAT_KIND_CHECKPOINTER 9 +#define PGSTAT_KIND_IO 10 +#define PGSTAT_KIND_SLRU 11 +#define PGSTAT_KIND_WAL 12 #define PGSTAT_KIND_BUILTIN_MIN PGSTAT_KIND_DATABASE #define PGSTAT_KIND_BUILTIN_MAX PGSTAT_KIND_WAL @@ -447,6 +449,7 @@ typedef struct PgStat_StatTabEntry PgStat_Counter blocks_fetched; PgStat_Counter blocks_hit; + PgStat_Counter blocks_written; TimestampTz last_vacuum_time; /* user initiated vacuum */ PgStat_Counter vacuum_count; @@ -458,6 +461,13 @@ typedef struct PgStat_StatTabEntry PgStat_Counter autoanalyze_count; } PgStat_StatTabEntry; +typedef struct PgStat_StatRelFileNodeEntry +{ + PgStat_Counter blocks_fetched; + PgStat_Counter blocks_hit; + PgStat_Counter blocks_written; +} PgStat_StatRelFileNodeEntry; + typedef struct PgStat_WalStats { PgStat_Counter wal_records; @@ -508,7 +518,7 @@ extern long pgstat_report_stat(bool force); extern void pgstat_force_next_flush(void); extern void pgstat_reset_counters(void); -extern void pgstat_reset(PgStat_Kind kind, Oid dboid, Oid objoid); +extern void pgstat_reset(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); extern void pgstat_reset_of_kind(PgStat_Kind kind); /* stats accessors */ @@ -517,7 +527,7 @@ extern TimestampTz pgstat_get_stat_snapshot_timestamp(bool *have_snapshot); /* helpers */ extern PgStat_Kind pgstat_get_kind_from_str(char *kind_str); -extern bool pgstat_have_entry(PgStat_Kind kind, Oid dboid, Oid objoid); +extern bool pgstat_have_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); /* @@ -626,6 +636,10 @@ extern void pgstat_report_analyze(Relation rel, PgStat_Counter livetuples, PgStat_Counter deadtuples, bool resetcounter); +extern void pgstat_report_relfilenode_blks_written(RelFileLocator locator); +extern void pgstat_report_relfilenode_buffer_read(Relation reln); +extern void pgstat_report_relfilenode_buffer_hit(Relation reln); + /* * If stats are enabled, but pending data hasn't been prepared yet, call * pgstat_assoc_relation() to do so. See its comment for why this is done @@ -685,6 +699,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info, void *recdata, uint32 len); extern PgStat_StatTabEntry *pgstat_fetch_stat_tabentry(Oid relid); +extern PgStat_StatRelFileNodeEntry *pgstat_fetch_stat_relfilenodeentry(Oid dboid, Oid spcOid, RelFileNumber relfile); extern PgStat_StatTabEntry *pgstat_fetch_stat_tabentry_ext(bool shared, Oid reloid); extern PgStat_TableStatus *find_tabstat_entry(Oid rel_id); diff --git a/src/include/utils/pgstat_internal.h b/src/include/utils/pgstat_internal.h index fb132e439d..47c448d5de 100644 --- a/src/include/utils/pgstat_internal.h +++ b/src/include/utils/pgstat_internal.h @@ -53,7 +53,8 @@ typedef struct PgStat_HashKey { PgStat_Kind kind; /* statistics entry kind */ Oid dboid; /* database ID. InvalidOid for shared objects. */ - Oid objoid; /* object ID, either table or function. */ + Oid objoid; /* object ID, either table or function or tablespace. */ + RelFileNumber relfile; /* relfilenumber for RelFileLocator. */ } PgStat_HashKey; /* @@ -390,6 +391,12 @@ typedef struct PgStatShared_Relation PgStat_StatTabEntry stats; } PgStatShared_Relation; +typedef struct PgStatShared_RelFileNode +{ + PgStatShared_Common header; + PgStat_StatRelFileNodeEntry stats; +} PgStatShared_RelFileNode; + typedef struct PgStatShared_Function { PgStatShared_Common header; @@ -528,6 +535,9 @@ static inline void *pgstat_get_entry_data(PgStat_Kind kind, PgStatShared_Common static inline void *pgstat_get_custom_shmem_data(PgStat_Kind kind); static inline void *pgstat_get_custom_snapshot_data(PgStat_Kind kind); +extern PgStat_SubXactStatus *pgStatXactStack; +extern void PgStat_RemoveRelFileNodeFromDroppedStats(PgStat_SubXactStatus *xact_state, RelFileLocator rlocator); + /* * Functions in pgstat.c @@ -544,10 +554,12 @@ extern void pgstat_assert_is_up(void); #endif extern void pgstat_delete_pending_entry(PgStat_EntryRef *entry_ref); -extern PgStat_EntryRef *pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, bool *created_entry); -extern PgStat_EntryRef *pgstat_fetch_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid); +extern PgStat_EntryRef *pgstat_prep_pending_entry(PgStat_Kind kind, Oid dboid, + Oid objoid, RelFileNumber relfile, + bool *created_entry); +extern PgStat_EntryRef *pgstat_fetch_pending_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); -extern void *pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid); +extern void *pgstat_fetch_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); extern void pgstat_snapshot_fixed(PgStat_Kind kind); @@ -619,6 +631,7 @@ extern void AtPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state); extern void PostPrepare_PgStat_Relations(PgStat_SubXactStatus *xact_state); extern bool pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); +extern bool pgstat_relfilenode_flush_cb(PgStat_EntryRef *entry_ref, bool nowait); extern void pgstat_relation_delete_pending_cb(PgStat_EntryRef *entry_ref); @@ -639,15 +652,16 @@ extern void pgstat_attach_shmem(void); extern void pgstat_detach_shmem(void); extern PgStat_EntryRef *pgstat_get_entry_ref(PgStat_Kind kind, Oid dboid, Oid objoid, - bool create, bool *created_entry); + RelFileNumber relfile, bool create, + bool *created_entry); extern bool pgstat_lock_entry(PgStat_EntryRef *entry_ref, bool nowait); extern bool pgstat_lock_entry_shared(PgStat_EntryRef *entry_ref, bool nowait); extern void pgstat_unlock_entry(PgStat_EntryRef *entry_ref); -extern bool pgstat_drop_entry(PgStat_Kind kind, Oid dboid, Oid objoid); +extern bool pgstat_drop_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); extern void pgstat_drop_all_entries(void); extern PgStat_EntryRef *pgstat_get_entry_ref_locked(PgStat_Kind kind, Oid dboid, Oid objoid, - bool nowait); -extern void pgstat_reset_entry(PgStat_Kind kind, Oid dboid, Oid objoid, TimestampTz ts); + RelFileNumber relfile, bool nowait); +extern void pgstat_reset_entry(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile, TimestampTz ts); extern void pgstat_reset_entries_of_kind(PgStat_Kind kind, TimestampTz ts); extern void pgstat_reset_matching_entries(bool (*do_reset) (PgStatShared_HashEntry *, Datum), Datum match_data, @@ -694,8 +708,8 @@ extern void pgstat_subscription_reset_timestamp_cb(PgStatShared_Common *header, */ extern PgStat_SubXactStatus *pgstat_get_xact_stack_level(int nest_level); -extern void pgstat_drop_transactional(PgStat_Kind kind, Oid dboid, Oid objoid); -extern void pgstat_create_transactional(PgStat_Kind kind, Oid dboid, Oid objoid); +extern void pgstat_drop_transactional(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); +extern void pgstat_create_transactional(PgStat_Kind kind, Oid dboid, Oid objoid, RelFileNumber relfile); /* diff --git a/src/test/recovery/t/029_stats_restart.pl b/src/test/recovery/t/029_stats_restart.pl index 93a7209f69..f9988b5028 100644 --- a/src/test/recovery/t/029_stats_restart.pl +++ b/src/test/recovery/t/029_stats_restart.pl @@ -40,10 +40,10 @@ trigger_funcrel_stat(); # verify stats objects exist my $sect = "initial"; -is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 't', "$sect: db stats do exist"); +is(have_stats('function', $dboid, $funcoid, 0), 't', "$sect: function stats do exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 't', "$sect: relation stats do exist"); # regular shutdown @@ -64,10 +64,10 @@ copy($og_stats, $statsfile) or die "Copy failed: $!"; $node->start; $sect = "copy"; -is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 't', "$sect: db stats do exist"); +is(have_stats('function', $dboid, $funcoid, 0), 't', "$sect: function stats do exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 't', "$sect: relation stats do exist"); $node->stop('immediate'); @@ -81,10 +81,10 @@ $node->start; # stats should have been discarded $sect = "post immediate"; -is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 'f', "$sect: db stats do not exist"); +is(have_stats('function', $dboid, $funcoid, 0), 'f', "$sect: function stats do exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 'f', "$sect: relation stats do not exist"); # get rid of backup statsfile @@ -95,10 +95,10 @@ unlink $statsfile or die "cannot unlink $statsfile $!"; trigger_funcrel_stat(); $sect = "post immediate, new"; -is(have_stats('database', $dboid, 0), 't', "$sect: db stats do exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 't', "$sect: db stats do exist"); +is(have_stats('function', $dboid, $funcoid, 0), 't', "$sect: function stats do exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 't', "$sect: relation stats do exist"); # regular shutdown @@ -114,10 +114,10 @@ $node->start; # no stats present due to invalid stats file $sect = "invalid_overwrite"; -is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 'f', "$sect: db stats do not exist"); +is(have_stats('function', $dboid, $funcoid, 0), 'f', "$sect: function stats do not exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 'f', "$sect: relation stats do not exist"); @@ -130,10 +130,10 @@ append_file($og_stats, "XYZ"); $node->start; $sect = "invalid_append"; -is(have_stats('database', $dboid, 0), 'f', "$sect: db stats do not exist"); -is(have_stats('function', $dboid, $funcoid), +is(have_stats('database', $dboid, 0, 0), 'f', "$sect: db stats do not exist"); +is(have_stats('function', $dboid, $funcoid, 0), 'f', "$sect: function stats do not exist"); -is(have_stats('relation', $dboid, $tableoid), +is(have_stats('relation', $dboid, $tableoid, 0), 'f', "$sect: relation stats do not exist"); @@ -292,10 +292,10 @@ sub trigger_funcrel_stat sub have_stats { - my ($kind, $dboid, $objoid) = @_; + my ($kind, $dboid, $objoid, $relfile) = @_; return $node->safe_psql($connect_db, - "SELECT pg_stat_have_stats('$kind', $dboid, $objoid)"); + "SELECT pg_stat_have_stats('$kind', $dboid, $objoid, $relfile)"); } sub overwrite_file diff --git a/src/test/recovery/t/030_stats_cleanup_replica.pl b/src/test/recovery/t/030_stats_cleanup_replica.pl index 74b516cc7c..317df24c4f 100644 --- a/src/test/recovery/t/030_stats_cleanup_replica.pl +++ b/src/test/recovery/t/030_stats_cleanup_replica.pl @@ -179,9 +179,9 @@ sub test_standby_func_tab_stats_status my %stats; $stats{rel} = $node_standby->safe_psql($connect_db, - "SELECT pg_stat_have_stats('relation', $dboid, $tableoid)"); + "SELECT pg_stat_have_stats('relation', $dboid, $tableoid, 0)"); $stats{func} = $node_standby->safe_psql($connect_db, - "SELECT pg_stat_have_stats('function', $dboid, $funcoid)"); + "SELECT pg_stat_have_stats('function', $dboid, $funcoid, 0)"); is_deeply(\%stats, \%expected, "$sect: standby stats as expected"); @@ -194,7 +194,7 @@ sub test_standby_db_stats_status my ($connect_db, $dboid, $present) = @_; is( $node_standby->safe_psql( - $connect_db, "SELECT pg_stat_have_stats('database', $dboid, 0)"), + $connect_db, "SELECT pg_stat_have_stats('database', $dboid, 0, 0)"), $present, "$sect: standby db stats as expected"); } diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out index 5201280669..234356a710 100644 --- a/src/test/regress/expected/rules.out +++ b/src/test/regress/expected/rules.out @@ -2335,6 +2335,11 @@ pg_statio_all_tables| SELECT c.oid AS relid, n.nspname AS schemaname, c.relname, (pg_stat_get_blocks_fetched(c.oid) - pg_stat_get_blocks_hit(c.oid)) AS heap_blks_read, + (pg_stat_get_blocks_written(c.oid) + pg_stat_get_relfilenode_blocks_written(d.oid, + CASE + WHEN (c.reltablespace <> (0)::oid) THEN c.reltablespace + ELSE d.dattablespace + END, c.relfilenode)) AS heap_blks_written, pg_stat_get_blocks_hit(c.oid) AS heap_blks_hit, i.idx_blks_read, i.idx_blks_hit, @@ -2342,7 +2347,8 @@ pg_statio_all_tables| SELECT c.oid AS relid, pg_stat_get_blocks_hit(t.oid) AS toast_blks_hit, x.idx_blks_read AS tidx_blks_read, x.idx_blks_hit AS tidx_blks_hit - FROM ((((pg_class c + FROM pg_database d, + ((((pg_class c LEFT JOIN pg_class t ON ((c.reltoastrelid = t.oid))) LEFT JOIN pg_namespace n ON ((n.oid = c.relnamespace))) LEFT JOIN LATERAL ( SELECT (sum((pg_stat_get_blocks_fetched(pg_index.indexrelid) - pg_stat_get_blocks_hit(pg_index.indexrelid))))::bigint AS idx_blks_read, @@ -2353,7 +2359,7 @@ pg_statio_all_tables| SELECT c.oid AS relid, (sum(pg_stat_get_blocks_hit(pg_index.indexrelid)))::bigint AS idx_blks_hit FROM pg_index WHERE (pg_index.indrelid = t.oid)) x ON (true)) - WHERE (c.relkind = ANY (ARRAY['r'::"char", 't'::"char", 'm'::"char"])); + WHERE ((c.relkind = ANY (ARRAY['r'::"char", 't'::"char", 'm'::"char"])) AND (d.datname = current_database())); pg_statio_sys_indexes| SELECT relid, indexrelid, schemaname, @@ -2374,6 +2380,7 @@ pg_statio_sys_tables| SELECT relid, schemaname, relname, heap_blks_read, + heap_blks_written, heap_blks_hit, idx_blks_read, idx_blks_hit, @@ -2403,6 +2410,7 @@ pg_statio_user_tables| SELECT relid, schemaname, relname, heap_blks_read, + heap_blks_written, heap_blks_hit, idx_blks_read, idx_blks_hit, diff --git a/src/test/regress/expected/stats.out b/src/test/regress/expected/stats.out index 6e08898b18..eff0c9372c 100644 --- a/src/test/regress/expected/stats.out +++ b/src/test/regress/expected/stats.out @@ -1111,23 +1111,23 @@ ROLLBACK; -- pg_stat_have_stats behavior ---- -- fixed-numbered stats exist -SELECT pg_stat_have_stats('bgwriter', 0, 0); +SELECT pg_stat_have_stats('bgwriter', 0, 0, 0); pg_stat_have_stats -------------------- t (1 row) -- unknown stats kinds error out -SELECT pg_stat_have_stats('zaphod', 0, 0); +SELECT pg_stat_have_stats('zaphod', 0, 0, 0); ERROR: invalid statistics kind: "zaphod" -- db stats have objoid 0 -SELECT pg_stat_have_stats('database', :dboid, 1); +SELECT pg_stat_have_stats('database', :dboid, 1, 0); pg_stat_have_stats -------------------- f (1 row) -SELECT pg_stat_have_stats('database', :dboid, 0); +SELECT pg_stat_have_stats('database', :dboid, 0, 0); pg_stat_have_stats -------------------- t @@ -1144,21 +1144,21 @@ select a from stats_test_tab1 where a = 3; 3 (1 row) -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t (1 row) -- pg_stat_have_stats returns false for dropped index with stats -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t (1 row) DROP index stats_test_idx1; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- f @@ -1174,14 +1174,14 @@ select a from stats_test_tab1 where a = 3; 3 (1 row) -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t (1 row) ROLLBACK; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- f @@ -1196,7 +1196,7 @@ select a from stats_test_tab1 where a = 3; 3 (1 row) -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t @@ -1204,7 +1204,7 @@ SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); REINDEX index CONCURRENTLY stats_test_idx1; -- false for previous oid -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- f @@ -1212,7 +1212,7 @@ SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); -- true for new oid SELECT 'stats_test_idx1'::regclass::oid AS stats_test_idx1_oid \gset -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t @@ -1220,7 +1220,7 @@ SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); -- pg_stat_have_stats returns true for a rolled back drop index with stats BEGIN; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t @@ -1228,7 +1228,7 @@ SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); DROP index stats_test_idx1; ROLLBACK; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); pg_stat_have_stats -------------------- t @@ -1513,7 +1513,7 @@ SELECT :io_sum_bulkwrite_strategy_extends_after > :io_sum_bulkwrite_strategy_ext (1 row) -- Test IO stats reset -SELECT pg_stat_have_stats('io', 0, 0); +SELECT pg_stat_have_stats('io', 0, 0, 0); pg_stat_have_stats -------------------- t diff --git a/src/test/regress/sql/stats.sql b/src/test/regress/sql/stats.sql index d8ac0d06f4..5a40779989 100644 --- a/src/test/regress/sql/stats.sql +++ b/src/test/regress/sql/stats.sql @@ -539,12 +539,12 @@ ROLLBACK; -- pg_stat_have_stats behavior ---- -- fixed-numbered stats exist -SELECT pg_stat_have_stats('bgwriter', 0, 0); +SELECT pg_stat_have_stats('bgwriter', 0, 0, 0); -- unknown stats kinds error out -SELECT pg_stat_have_stats('zaphod', 0, 0); +SELECT pg_stat_have_stats('zaphod', 0, 0, 0); -- db stats have objoid 0 -SELECT pg_stat_have_stats('database', :dboid, 1); -SELECT pg_stat_have_stats('database', :dboid, 0); +SELECT pg_stat_have_stats('database', :dboid, 1, 0); +SELECT pg_stat_have_stats('database', :dboid, 0, 0); -- pg_stat_have_stats returns true for committed index creation CREATE table stats_test_tab1 as select generate_series(1,10) a; @@ -552,40 +552,40 @@ CREATE index stats_test_idx1 on stats_test_tab1(a); SELECT 'stats_test_idx1'::regclass::oid AS stats_test_idx1_oid \gset SET enable_seqscan TO off; select a from stats_test_tab1 where a = 3; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- pg_stat_have_stats returns false for dropped index with stats -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); DROP index stats_test_idx1; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- pg_stat_have_stats returns false for rolled back index creation BEGIN; CREATE index stats_test_idx1 on stats_test_tab1(a); SELECT 'stats_test_idx1'::regclass::oid AS stats_test_idx1_oid \gset select a from stats_test_tab1 where a = 3; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); ROLLBACK; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- pg_stat_have_stats returns true for reindex CONCURRENTLY CREATE index stats_test_idx1 on stats_test_tab1(a); SELECT 'stats_test_idx1'::regclass::oid AS stats_test_idx1_oid \gset select a from stats_test_tab1 where a = 3; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); REINDEX index CONCURRENTLY stats_test_idx1; -- false for previous oid -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- true for new oid SELECT 'stats_test_idx1'::regclass::oid AS stats_test_idx1_oid \gset -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- pg_stat_have_stats returns true for a rolled back drop index with stats BEGIN; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); DROP index stats_test_idx1; ROLLBACK; -SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid); +SELECT pg_stat_have_stats('relation', :dboid, :stats_test_idx1_oid, 0); -- put enable_seqscan back to on SET enable_seqscan TO on; @@ -759,7 +759,7 @@ SELECT sum(extends) AS io_sum_bulkwrite_strategy_extends_after SELECT :io_sum_bulkwrite_strategy_extends_after > :io_sum_bulkwrite_strategy_extends_before; -- Test IO stats reset -SELECT pg_stat_have_stats('io', 0, 0); +SELECT pg_stat_have_stats('io', 0, 0, 0); SELECT sum(evictions) + sum(reuses) + sum(extends) + sum(fsyncs) + sum(reads) + sum(writes) + sum(writebacks) + sum(hits) AS io_stats_pre_reset FROM pg_stat_io \gset SELECT pg_stat_reset_shared('io'); diff --git a/src/test/subscription/t/026_stats.pl b/src/test/subscription/t/026_stats.pl index fb3e5629b3..1f4ae5efd5 100644 --- a/src/test/subscription/t/026_stats.pl +++ b/src/test/subscription/t/026_stats.pl @@ -263,7 +263,7 @@ $node_subscriber->safe_psql($db, qq(DROP SUBSCRIPTION $sub1_name)); # Subscription stats for sub1 should be gone is( $node_subscriber->safe_psql( - $db, qq(SELECT pg_stat_have_stats('subscription', 0, $sub1_oid))), + $db, qq(SELECT pg_stat_have_stats('subscription', 0, $sub1_oid, 0))), qq(f), qq(Subscription stats for subscription '$sub1_name' should be removed.)); @@ -282,7 +282,7 @@ DROP SUBSCRIPTION $sub2_name; # Subscription stats for sub2 should be gone is( $node_subscriber->safe_psql( - $db, qq(SELECT pg_stat_have_stats('subscription', 0, $sub2_oid))), + $db, qq(SELECT pg_stat_have_stats('subscription', 0, $sub2_oid, 0))), qq(f), qq(Subscription stats for subscription '$sub2_name' should be removed.)); diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 6e6b7c2711..7b6b413c03 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2123,6 +2123,7 @@ PgStatShared_InjectionPoint PgStatShared_InjectionPointFixed PgStatShared_IO PgStatShared_Relation +PgStatShared_RelFileNode PgStatShared_ReplSlot PgStatShared_SLRU PgStatShared_Subscription -- 2.34.1 --fLO+MFZDcAUMk1It-- ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 178+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 178+ messages in thread
end of thread, other threads:[~2026-04-06 22:07 UTC | newest] Thread overview: 178+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2022-06-17 01:33 [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. Andres Freund <[email protected]> 2023-11-16 02:30 [PATCH v3] Provide relfilenode statistics Bertrand Drouvot <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[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