agora inbox for [email protected]help / color / mirror / Atom feed
[PATCH 2/2] Publish list of tables being repacked in shared memory 102+ messages / 2 participants [nested] [flat]
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH 2/2] Publish list of tables being repacked in shared memory @ 2026-04-07 20:29 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Álvaro Herrera @ 2026-04-07 20:29 UTC (permalink / raw) Use it in autovacuum to skip processing tables that are being repacked. This is mostly to avoid repeated attempts to process such tables, which would fail due to the special deadlock checker behavior for repack. Author: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/[email protected] --- src/backend/commands/repack.c | 195 ++++++++++++++++-- src/backend/postmaster/autovacuum.c | 20 ++ .../utils/activity/wait_event_names.txt | 1 + src/include/commands/repack.h | 2 + src/include/storage/lwlocklist.h | 2 +- src/include/storage/subsystemlist.h | 1 + src/tools/pgindent/typedefs.list | 3 + 7 files changed, 210 insertions(+), 14 deletions(-) diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c index a5f5df77291..ee7072dce6a 100644 --- a/src/backend/commands/repack.c +++ b/src/backend/commands/repack.c @@ -63,9 +63,11 @@ #include "optimizer/optimizer.h" #include "pgstat.h" #include "storage/bufmgr.h" +#include "storage/ipc.h" #include "storage/lmgr.h" #include "storage/predicate.h" #include "storage/proc.h" +#include "storage/subsystems.h" #include "utils/acl.h" #include "utils/fmgroids.h" #include "utils/guc.h" @@ -79,6 +81,32 @@ #include "utils/syscache.h" #include "utils/wait_event_types.h" + +/* Shared memory layout for REPACK */ +typedef struct RepackWorkerInfo +{ + bool ri_in_use; + pid_t ri_backendpid; + Oid ri_dbid; + Oid ri_relid; + Oid ri_toastrelid; +} RepackWorkerInfo; + +typedef struct +{ + bool re_useless; + RepackWorkerInfo re_workerinfo[FLEXIBLE_ARRAY_MEMBER]; +} RepackShmemStruct; + +static RepackShmemStruct *RepackShmem; + +typedef struct RepackCleanupContext +{ + bool concurrent; + int workerindex; +} RepackCleanupContext; + + /* * This struct is used to pass around the information on tables to be * clustered. We need this so we can make a list of them when invoked without @@ -90,6 +118,7 @@ typedef struct Oid indexOid; } RelToCluster; + /* * The first file exported by the decoding worker must contain a snapshot, the * following ones contain the data changes. @@ -166,6 +195,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd, MemoryContext permcxt); static bool repack_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid); +static void RepackCleanup(RepackCleanupContext *context); +static void RepackCleanupCb(int code, Datum arg); +static void RepackShmemRequest(void *arg); +static void RepackShmemInit(void *arg); static void apply_concurrent_changes(BufFile *file, ChangeContext *chgcxt); static void apply_concurrent_insert(Relation rel, TupleTableSlot *slot, @@ -210,6 +243,11 @@ static void ProcessRepackMessage(StringInfo msg); static const char *RepackCommandAsString(RepackCommand cmd); +const ShmemCallbacks RepackShmemCallbacks = { + .request_fn = RepackShmemRequest, + .init_fn = RepackShmemInit, +}; + /* * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we @@ -514,6 +552,7 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, Oid tableOid = RelationGetRelid(OldHeap); Relation index; LOCKMODE lmode; + RepackCleanupContext context; Oid save_userid; int save_sec_context; int save_nestlevel; @@ -660,24 +699,43 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, TransferPredicateLocksToHeapRelation(OldHeap); /* rebuild_relation does all the dirty work */ - PG_TRY(); - { - rebuild_relation(OldHeap, index, verbose, ident_idx); - } - PG_FINALLY(); + context.concurrent = concurrent; + + PG_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); { if (concurrent) { - /* - * Since during normal operation the worker was already asked to - * exit, stopping it explicitly is especially important on ERROR. - * However it still seems a good practice to make sure that the - * worker never survives the REPACK command. - */ - stop_repack_decoding_worker(); + bool freefound = false; + + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *worker; + + if (RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + freefound = true; + worker = &RepackShmem->re_workerinfo[i]; + context.workerindex = i; + + worker->ri_in_use = true; + worker->ri_backendpid = MyProcPid; + worker->ri_dbid = MyDatabaseId; + worker->ri_relid = RelationGetRelid(OldHeap); + worker->ri_toastrelid = OldHeap->rd_rel->reltoastrelid; + break; + } + if (!freefound) + elog(ERROR, "could not find free repack entry"); + LWLockRelease(RepackLock); } + + rebuild_relation(OldHeap, index, verbose, ident_idx); } - PG_END_TRY(); + PG_END_ENSURE_ERROR_CLEANUP(RepackCleanupCb, PointerGetDatum(&context)); + + RepackCleanup(&context); /* rebuild_relation closes OldHeap, and index if valid */ @@ -691,6 +749,117 @@ out: pgstat_progress_end_command(); } +/* + * Return whether any backend is running concurrent REPACK on the given table + * (which could be a toast table). + */ +bool +is_table_under_repack(Oid databaseId, Oid relid) +{ + bool retval = false; + + LWLockAcquire(RepackLock, LW_SHARED); + for (int i = 0; i < max_repack_replication_slots; i++) + { + RepackWorkerInfo *rworker; + + if (!RepackShmem->re_workerinfo[i].ri_in_use) + continue; + + rworker = &RepackShmem->re_workerinfo[i]; + if (rworker->ri_dbid == MyDatabaseId && + (rworker->ri_relid == relid || + rworker->ri_toastrelid == relid)) + retval = true; + } + LWLockRelease(RepackLock); + + return retval; +} + +/* + * Remove ourselves from the workerinfo array. + */ +static void +RepackCleanup(RepackCleanupContext *context) +{ + if (context->concurrent) + { + RepackWorkerInfo *worker; + + /* + * The worker would normally terminate on its own when the work is + * done, but make sure we signal it just in case. + */ + stop_repack_decoding_worker(); + + /* + * also, make sure we stop advertising the relation we were repacking, + * so that autovacuum reverts to handling it normally. + */ + LWLockAcquire(RepackLock, LW_EXCLUSIVE); + + worker = &RepackShmem->re_workerinfo[context->workerindex]; + Assert(worker->ri_backendpid == MyProcPid); + worker->ri_in_use = false; + worker->ri_backendpid = 0; + worker->ri_dbid = InvalidOid; + worker->ri_relid = InvalidOid; + worker->ri_toastrelid = InvalidOid; + LWLockRelease(RepackLock); + } +} + +/* + * RepackCleanup wrapped as an on_shmem_exit callback function + */ +static void +RepackCleanupCb(int code, Datum arg) +{ + RepackCleanup((RepackCleanupContext *) DatumGetPointer(arg)); +} + +/* + * RepackShmemRequest + * Register shared memory space needed for repack + */ +static void +RepackShmemRequest(void *arg) +{ + Size size; + + /* + * Need the fixed struct and the array of RepackWorkerInfo. + */ + size = sizeof(RepackShmemStruct); + size = MAXALIGN(size); + size = add_size(size, mul_size(max_repack_replication_slots, + sizeof(RepackWorkerInfo))); + + ShmemRequestStruct(.name = "Repack Data", + .size = size, + .ptr = (void **) &RepackShmem, + ); +} + +static void +RepackShmemInit(void *arg) +{ + RepackWorkerInfo *reinfo; + + reinfo = (RepackWorkerInfo *) ((char *) RepackShmem + + MAXALIGN(sizeof(RepackShmemStruct))); + + for (int i = 0; i < max_repack_replication_slots; i++) + { + reinfo[i].ri_in_use = false; + reinfo[i].ri_backendpid = 0; + reinfo[i].ri_dbid = InvalidOid; + reinfo[i].ri_relid = InvalidOid; + reinfo[i].ri_toastrelid = InvalidOid; + } +} + /* * Check if the table (and its index) still meets the requirements of * cluster_rel(). diff --git a/src/backend/postmaster/autovacuum.c b/src/backend/postmaster/autovacuum.c index bd626a16363..080c64ea3c8 100644 --- a/src/backend/postmaster/autovacuum.c +++ b/src/backend/postmaster/autovacuum.c @@ -78,6 +78,7 @@ #include "catalog/namespace.h" #include "catalog/pg_database.h" #include "catalog/pg_namespace.h" +#include "commands/repack.h" #include "commands/vacuum.h" #include "common/int.h" #include "funcapi.h" @@ -2422,6 +2423,25 @@ do_autovacuum(void) } } LWLockRelease(AutovacuumLock); + + /* + * Similarly, if the table is being processed by concurrent repack, + * skip it (but make a note of that). We wouldn't be able to acquire + * its lock anyway. + */ + if (!skipit) + { + MemoryContextSwitchTo(PortalContext); + + skipit = is_table_under_repack(MyDatabaseId, relid); + if (skipit) + ereport(LOG, + errmsg("skipping table \"%s.%s.%s\" because it's being repacked in concurrent mode", + get_database_name(MyDatabaseId), + get_namespace_name(get_rel_namespace(relid)), + get_rel_name(relid))); + } + if (skipit) { LWLockRelease(AutovacuumScheduleLock); diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt index 7bda5298558..e206304f204 100644 --- a/src/backend/utils/activity/wait_event_names.txt +++ b/src/backend/utils/activity/wait_event_names.txt @@ -332,6 +332,7 @@ SInvalWrite "Waiting to add a message to the shared catalog invalidation queue." WALBufMapping "Waiting to replace a page in WAL buffers." WALWrite "Waiting for WAL buffers to be written to disk." ControlFile "Waiting to read or update the <filename>pg_control</filename> file or create a new WAL file." +Repack "Waiting to read or update tables in process by concurrent repack." MultiXactGen "Waiting to read or update shared multixact state." RelCacheInit "Waiting to read or update a <filename>pg_internal.init</filename> relation cache initialization file." CheckpointerComm "Waiting to manage fsync requests." diff --git a/src/include/commands/repack.h b/src/include/commands/repack.h index fd16e74b179..be7d38b5fae 100644 --- a/src/include/commands/repack.h +++ b/src/include/commands/repack.h @@ -42,6 +42,8 @@ extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel); extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid, ClusterParams *params, bool isTopLevel); +extern bool is_table_under_repack(Oid databaseId, Oid relid); + extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal); diff --git a/src/include/storage/lwlocklist.h b/src/include/storage/lwlocklist.h index af8553bcb6c..3f08f4a15d4 100644 --- a/src/include/storage/lwlocklist.h +++ b/src/include/storage/lwlocklist.h @@ -41,7 +41,7 @@ PG_LWLOCK(6, SInvalWrite) PG_LWLOCK(7, WALBufMapping) PG_LWLOCK(8, WALWrite) PG_LWLOCK(9, ControlFile) -/* 10 was CheckpointLock */ +PG_LWLOCK(10, Repack) /* 11 was XactSLRULock */ /* 12 was SubtransSLRULock */ PG_LWLOCK(13, MultiXactGen) diff --git a/src/include/storage/subsystemlist.h b/src/include/storage/subsystemlist.h index 9ad619080be..4e683b8b0a8 100644 --- a/src/include/storage/subsystemlist.h +++ b/src/include/storage/subsystemlist.h @@ -72,6 +72,7 @@ PG_SHMEM_SUBSYSTEM(WalSummarizerShmemCallbacks) PG_SHMEM_SUBSYSTEM(PgArchShmemCallbacks) PG_SHMEM_SUBSYSTEM(ApplyLauncherShmemCallbacks) PG_SHMEM_SUBSYSTEM(SlotSyncShmemCallbacks) +PG_SHMEM_SUBSYSTEM(RepackShmemCallbacks) /* other modules that need some shared memory space */ PG_SHMEM_SUBSYSTEM(BTreeShmemCallbacks) diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 637c669a146..d019e03aaf1 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2639,9 +2639,12 @@ ReorderBufferTupleCidEnt ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple +RepackCleanupContext RepackCommand RepackDecodingState +RepackShmemStruct RepackStmt +RepackWorkerInfo ReparameterizeForeignPathByChild_function ReplOriginId ReplOriginXactState -- 2.47.3 --kdrcpfmkbkc4lqhu-- ^ permalink raw reply [nested|flat] 102+ messages in thread
* [PATCH v20 3/3] Add Assert guard to detect permission check before lock regressions @ 2026-04-27 14:23 Bertrand Drouvot <[email protected]> 0 siblings, 0 replies; 102+ messages in thread From: Bertrand Drouvot @ 2026-04-27 14:23 UTC (permalink / raw) Add instrumentation under USE_ASSERT_CHECKING to detect cases where object_aclcheck() is called on a referenced object before a lock is held on it, which would widen the TOCTOU window between the permission check and the dependency recording. In recordMultipleDependencies() and changeDependencyFor(), for each referenced object that had a permission check earlier in the same statement, assert that a lock is already held. This catches any future code that adds a permission check on a referenced object without first acquiring a lock. The tracking records each (classId, objectId, mode) from object_aclcheck() and pg_class_aclcheck(), filtering to only dependency-relevant ACL modes (ACL_CREATE, ACL_USAGE on non namespaces, ACL_EXECUTE, ACL_TRIGGER, ACL_REFERENCES). Tracking resets at each ProcessUtility() call. The tracking array is capped at 1024 entries per statement, which is sufficient for any practical DDL command. Author: Bertrand Drouvot <[email protected]> Reviewed-by: Discussion: https://postgr.es/m/[email protected] --- src/backend/catalog/aclchk.c | 32 ++++++++++++ src/backend/catalog/pg_depend.c | 52 ++++++++++++++++++++ src/backend/tcop/utility.c | 3 ++ src/include/catalog/aclcheck_track.h | 73 ++++++++++++++++++++++++++++ src/tools/pgindent/typedefs.list | 1 + 5 files changed, 161 insertions(+) 51.5% src/backend/catalog/ 46.7% src/include/catalog/ diff --git a/src/backend/catalog/aclchk.c b/src/backend/catalog/aclchk.c index 67424fe3b0c..3c3b35910e1 100644 --- a/src/backend/catalog/aclchk.c +++ b/src/backend/catalog/aclchk.c @@ -76,6 +76,7 @@ #include "parser/parse_type.h" #include "storage/lmgr.h" #include "utils/acl.h" +#include "catalog/aclcheck_track.h" #include "utils/aclchk_internal.h" #include "utils/builtins.h" #include "utils/fmgroids.h" @@ -3890,6 +3891,8 @@ object_aclcheck_ext(Oid classid, Oid objectid, Oid roleid, AclMode mode, bool *is_missing) { + aclcheck_track_record(classid, objectid, mode); + if (object_aclmask_ext(classid, objectid, roleid, mode, ACLMASK_ANY, is_missing) != 0) return ACLCHECK_OK; @@ -4092,6 +4095,8 @@ AclResult pg_class_aclcheck_ext(Oid table_oid, Oid roleid, AclMode mode, bool *is_missing) { + aclcheck_track_record(RelationRelationId, table_oid, mode); + if (pg_class_aclmask_ext(table_oid, roleid, mode, ACLMASK_ANY, is_missing) != 0) return ACLCHECK_OK; @@ -5035,3 +5040,30 @@ RemoveRoleFromInitPriv(Oid roleid, Oid classid, Oid objid, int32 objsubid) table_close(rel, RowExclusiveLock); } + +/* + * Instrumentation to detect permission check before lock regressions. + * Only used in assert-enabled builds. + */ +#ifdef USE_ASSERT_CHECKING +AclCheckEntry aclcheck_tracked[ACLCHECK_TRACK_MAX]; +int aclcheck_tracked_count = 0; + +void +aclcheck_track_reset(void) +{ + aclcheck_tracked_count = 0; +} + +bool +aclcheck_track_was_checked(Oid classId, Oid objectId) +{ + for (int i = 0; i < aclcheck_tracked_count; i++) + { + if (aclcheck_tracked[i].classId == classId && + aclcheck_tracked[i].objectId == objectId) + return true; + } + return false; +} +#endif /* USE_ASSERT_CHECKING */ diff --git a/src/backend/catalog/pg_depend.c b/src/backend/catalog/pg_depend.c index 5618e3d26fa..28fa99b2d42 100644 --- a/src/backend/catalog/pg_depend.c +++ b/src/backend/catalog/pg_depend.c @@ -18,6 +18,7 @@ #include "access/htup_details.h" #include "access/table.h" #include "catalog/catalog.h" +#include "catalog/aclcheck_track.h" #include "catalog/dependency.h" #include "catalog/indexing.h" #include "catalog/pg_constraint.h" @@ -27,6 +28,8 @@ #include "catalog/partition.h" #include "commands/extension.h" #include "miscadmin.h" +#include "storage/lmgr.h" +#include "storage/lock.h" #include "utils/fmgroids.h" #include "utils/lsyscache.h" #include "utils/rel.h" @@ -107,6 +110,31 @@ recordMultipleDependencies(const ObjectAddress *depender, if (isObjectPinned(referenced)) continue; +#ifdef USE_ASSERT_CHECKING + /* + * If this referenced object had a permission check earlier in this + * statement, assert that a lock is already held on it. This ensures + * callers acquired the lock before calling object_aclcheck(), not + * after. The latter would widen the TOCTOU window between the + * permission check and the dependency recording. + */ + if (aclcheck_track_was_checked(referenced->classId, referenced->objectId)) + { + if (referenced->classId == RelationRelationId) + Assert(CheckRelationOidLockedByMe(referenced->objectId, + AccessShareLock, true)); + else + { + LOCKTAG tag; + + SET_LOCKTAG_OBJECT(tag, MyDatabaseId, + referenced->classId, + referenced->objectId, 0); + Assert(LockHeldByMe(&tag, AccessShareLock, true)); + } + } +#endif + /* * Acquire a lock and check object still exists while recording the * dependency. @@ -511,6 +539,30 @@ changeDependencyFor(Oid classId, Oid objectId, return 1; } +#ifdef USE_ASSERT_CHECKING + /* + * If this referenced object had a permission check earlier in this + * statement, assert that a lock is already held on it. This ensures + * callers acquired the lock before calling object_aclcheck(), not after. + * The latter would widen the TOCTOU window between the permission check and + * the dependency recording. + */ + if (aclcheck_track_was_checked(objAddr.classId, objAddr.objectId)) + { + if (objAddr.classId == RelationRelationId) + Assert(CheckRelationOidLockedByMe(objAddr.objectId, + AccessShareLock, true)); + else + { + LOCKTAG tag; + + SET_LOCKTAG_OBJECT(tag, MyDatabaseId, + objAddr.classId, + objAddr.objectId, 0); + Assert(LockHeldByMe(&tag, AccessShareLock, true)); + } + } +#endif /* * Acquire a lock and check object still exists while changing the * dependency. diff --git a/src/backend/tcop/utility.c b/src/backend/tcop/utility.c index 73a56f1df1d..0c4d382bd37 100644 --- a/src/backend/tcop/utility.c +++ b/src/backend/tcop/utility.c @@ -21,6 +21,7 @@ #include "access/xact.h" #include "access/xlog.h" #include "catalog/namespace.h" +#include "catalog/aclcheck_track.h" #include "catalog/pg_authid.h" #include "catalog/pg_inherits.h" #include "catalog/toasting.h" @@ -515,6 +516,8 @@ ProcessUtility(PlannedStmt *pstmt, Assert(queryString != NULL); /* required as of 8.4 */ Assert(qc == NULL || qc->commandTag == CMDTAG_UNKNOWN); + aclcheck_track_reset(); + /* * We provide a function hook variable that lets loadable plugins get * control when ProcessUtility is called. Such a plugin would normally diff --git a/src/include/catalog/aclcheck_track.h b/src/include/catalog/aclcheck_track.h new file mode 100644 index 00000000000..5825d7398d3 --- /dev/null +++ b/src/include/catalog/aclcheck_track.h @@ -0,0 +1,73 @@ +/*------------------------------------------------------------------------- + * + * aclcheck_track.h + * Instrumentation to detect permission check before lock via Assert in + * recordMultipleDependencies() and changeDependencyFor(). + * + * Only active in USE_ASSERT_CHECKING builds. + * + * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * src/include/catalog/aclcheck_track.h + * + *------------------------------------------------------------------------- + */ +#ifndef ACLCHECK_TRACK_H +#define ACLCHECK_TRACK_H + +#ifdef USE_ASSERT_CHECKING + +#include "catalog/pg_namespace.h" + +#define ACLCHECK_TRACK_MAX 1024 + +typedef struct AclCheckEntry +{ + Oid classId; + Oid objectId; +} AclCheckEntry; + +extern AclCheckEntry aclcheck_tracked[]; +extern int aclcheck_tracked_count; + +extern void aclcheck_track_reset(void); +extern bool aclcheck_track_was_checked(Oid classId, Oid objectId); + +/* + * Only record aclchecks that are dependency-relevant: + * - ACL_CREATE on any object (creating something in a container) + * - ACL_USAGE on non-namespace objects (using a type, language, server) + * - ACL_EXECUTE on functions + * - ACL_TRIGGER, ACL_REFERENCES on relations + * + * Skip ACL_USAGE on namespaces — that's name resolution, not dependency. + */ +static inline void +aclcheck_track_record(Oid classId, Oid objectId, AclMode mode) +{ + /* Skip ACL_USAGE on namespaces (name resolution, not dependency) */ + if (classId == NamespaceRelationId && !(mode & ACL_CREATE)) + return; + + /* Only track dependency-relevant modes */ + if (!(mode & (ACL_CREATE | ACL_USAGE | ACL_EXECUTE | ACL_TRIGGER | ACL_REFERENCES))) + return; + + if (aclcheck_tracked_count < ACLCHECK_TRACK_MAX) + { + aclcheck_tracked[aclcheck_tracked_count].classId = classId; + aclcheck_tracked[aclcheck_tracked_count].objectId = objectId; + aclcheck_tracked_count++; + } +} + +#else /* !USE_ASSERT_CHECKING */ + +#define aclcheck_track_reset() ((void) 0) +#define aclcheck_track_record(c, o, m) ((void) 0) +#define aclcheck_track_was_checked(c, o) (false) + +#endif /* USE_ASSERT_CHECKING */ + +#endif /* ACLCHECK_TRACK_H */ diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list index 9f1dd55213d..0a8a2772b23 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -20,6 +20,7 @@ AbsoluteTime AccessMethodInfo AccessPriv Acl +AclCheckEntry AclItem AclMaskHow AclMode -- 2.34.1 --zB/Wek73NRZcta1g-- ^ permalink raw reply [nested|flat] 102+ messages in thread
end of thread, other threads:[~2026-04-27 14:23 UTC | newest] Thread overview: 102+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-07 20:29 [PATCH 2/2] Publish list of tables being repacked in shared memory Álvaro Herrera <[email protected]> 2026-04-27 14:23 [PATCH v20 3/3] Add Assert guard to detect permission check before lock regressions Bertrand Drouvot <[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