Lock before setting relhassubclass on RELKIND_PARTITIONED_INDEX.

Commit 5b562644fec696977df4a82790064e8287927891 added a comment that
SetRelationHasSubclass() callers must hold this lock.  When commit
17f206fbc824d2b4b14480199ca9ff7dea417eda extended use of this column to
partitioned indexes, it didn't take the lock.  As the latter commit
message mentioned, we currently never reset a partitioned index to
relhassubclass=f.  That largely avoids harm from the lock omission.  The
cause for fixing this now is to unblock introducing a rule about locks
required to heap_update() a pg_class row.  This might cause more
deadlocks.  It gives minor user-visible benefits:

- If an ALTER INDEX SET TABLESPACE runs concurrently with ALTER TABLE
  ATTACH PARTITION or CREATE PARTITION OF, one transaction blocks
  instead of failing with "tuple concurrently updated".  (Many cases of
  DDL concurrency still fail that way.)

- Match ALTER INDEX ATTACH PARTITION in choosing to lock the index.

While not user-visible today, we'll need this if we ever make something
set the flag to false for a partitioned index, like ANALYZE does today
for tables.  Back-patch to v12 (all supported versions), the plan for
the commit relying on the new rule.  In back branches, add
LockOrStrongerHeldByMe() instead of adding a LockHeldByMe() parameter.

Reviewed (in an earlier version) by Robert Haas.

Discussion: https://postgr.es/m/20240611024525[email protected]

diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c
index 9e39d949b5eb22ebe1399499eb76be1e1579a012..cfa30ac38cf829fbd48cdaad815b0d4a4b333a13 100644 (file)
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -1014,6 +1014,7 @@ index_create(Relation heapRelation,
    if (OidIsValid(parentIndexRelid))
    {
        StoreSingleInheritance(indexRelationId, parentIndexRelid, 1);
+       LockRelationOid(parentIndexRelid, ShareUpdateExclusiveLock);
        SetRelationHasSubclass(parentIndexRelid, true);
    }
 

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 075adced770ce21ba72cdfe3f12b594f80cd8949..1d66376927fbbf240fcbbc4f26e47e86ddce1387 100644 (file)
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -4245,7 +4245,10 @@ IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
 
    /* set relhassubclass if an index partition has been added to the parent */
    if (OidIsValid(parentOid))
+   {
+       LockRelationOid(parentOid, ShareUpdateExclusiveLock);
        SetRelationHasSubclass(parentOid, true);
+   }
 
    /* set relispartition correctly on the partition */
    update_relispartition(partRelid, OidIsValid(parentOid));

diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
index 9ec5d692d9b8b4b3ead54a9ad28a34fa15a5ab30..6448c596a85364e98caf026c223729583c27a6a8 100644 (file)
--- a/src/backend/commands/tablecmds.c
+++ b/src/backend/commands/tablecmds.c
@@ -3196,8 +3196,15 @@ findAttrByName(const char *attributeName, List *schema)
  * SetRelationHasSubclass
  *     Set the value of the relation's relhassubclass field in pg_class.
  *
- * NOTE: caller must be holding an appropriate lock on the relation.
- * ShareUpdateExclusiveLock is sufficient.
+ * It's always safe to set this field to true, because all SQL commands are
+ * ready to see true and then find no children.  On the other hand, commands
+ * generally assume zero children if this is false.
+ *
+ * Caller must hold any self-exclusive lock until end of transaction.  If the
+ * new value is false, caller must have acquired that lock before reading the
+ * evidence that justified the false value.  That way, it properly waits if
+ * another backend is simultaneously concluding no need to change the tuple
+ * (new and old values are true).
  *
  * NOTE: an important side-effect of this operation is that an SI invalidation
  * message is sent out to all backends --- including me --- causing plans
@@ -3212,6 +3219,11 @@ SetRelationHasSubclass(Oid relationId, bool relhassubclass)
    HeapTuple   tuple;
    Form_pg_class classtuple;
 
+   Assert(CheckRelationOidLockedByMe(relationId,
+                                     ShareUpdateExclusiveLock, false) ||
+          CheckRelationOidLockedByMe(relationId,
+                                     ShareRowExclusiveLock, true));
+
    /*
     * Fetch a modifiable copy of the tuple, modify it, update pg_class.
     */

diff --git a/src/backend/storage/lmgr/lmgr.c b/src/backend/storage/lmgr/lmgr.c
index b717ba2a4b9ad9e07f9c66edb02317cc51a46980..b54f3fb5382f94c38ef7500fb662d9ce71e35443 100644 (file)
--- a/src/backend/storage/lmgr/lmgr.c
+++ b/src/backend/storage/lmgr/lmgr.c
@@ -308,32 +308,26 @@ CheckRelationLockedByMe(Relation relation, LOCKMODE lockmode, bool orstronger)
                         relation->rd_lockInfo.lockRelId.dbId,
                         relation->rd_lockInfo.lockRelId.relId);
 
-   if (LockHeldByMe(&tag, lockmode))
-       return true;
+   return (orstronger ?
+           LockOrStrongerHeldByMe(&tag, lockmode) :
+           LockHeldByMe(&tag, lockmode));
+}
 
-   if (orstronger)
-   {
-       LOCKMODE    slockmode;
+/*
+ *     CheckRelationOidLockedByMe
+ *
+ * Like the above, but takes an OID as argument.
+ */
+bool
+CheckRelationOidLockedByMe(Oid relid, LOCKMODE lockmode, bool orstronger)
+{
+   LOCKTAG     tag;
 
-       for (slockmode = lockmode + 1;
-            slockmode <= MaxLockMode;
-            slockmode++)
-       {
-           if (LockHeldByMe(&tag, slockmode))
-           {
-#ifdef NOT_USED
-               /* Sometimes this might be useful for debugging purposes */
-               elog(WARNING, "lock mode %s substituted for %s on relation %s",
-                    GetLockmodeName(tag.locktag_lockmethodid, slockmode),
-                    GetLockmodeName(tag.locktag_lockmethodid, lockmode),
-                    RelationGetRelationName(relation));
-#endif
-               return true;
-           }
-       }
-   }
+   SetLocktagRelationOid(&tag, relid);
 
-   return false;
+   return (orstronger ?
+           LockOrStrongerHeldByMe(&tag, lockmode) :
+           LockHeldByMe(&tag, lockmode));
 }
 
 /*

diff --git a/src/backend/storage/lmgr/lock.c b/src/backend/storage/lmgr/lock.c
index a1bf6ad792a3e9753fa7bc0c8d999d2102413707..aec56d5bb3b5844a7aaf2fd63a2e0219121e7a8c 100644 (file)
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -578,11 +578,17 @@ DoLockModesConflict(LOCKMODE mode1, LOCKMODE mode2)
 }
 
 /*
- * LockHeldByMe -- test whether lock 'locktag' is held with mode 'lockmode'
- *     by the current transaction
+ * LockHeldByMeExtended -- test whether lock 'locktag' is held by the current
+ *     transaction
+ *
+ * Returns true if current transaction holds a lock on 'tag' of mode
+ * 'lockmode'.  If 'orstronger' is true, a stronger lockmode is also OK.
+ * ("Stronger" is defined as "numerically higher", which is a bit
+ * semantically dubious but is OK for the purposes we use this for.)
  */
-bool
-LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
+static bool
+LockHeldByMeExtended(const LOCKTAG *locktag,
+                    LOCKMODE lockmode, bool orstronger)
 {
    LOCALLOCKTAG localtag;
    LOCALLOCK  *locallock;
@@ -598,7 +604,35 @@ LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
                                          (void *) &localtag,
                                          HASH_FIND, NULL);
 
-   return (locallock && locallock->nLocks > 0);
+   if (locallock && locallock->nLocks > 0)
+       return true;
+
+   if (orstronger)
+   {
+       LOCKMODE    slockmode;
+
+       for (slockmode = lockmode + 1;
+            slockmode <= MaxLockMode;
+            slockmode++)
+       {
+           if (LockHeldByMeExtended(locktag, slockmode, false))
+               return true;
+       }
+   }
+
+   return false;
+}
+
+bool
+LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
+{
+   return LockHeldByMeExtended(locktag, lockmode, false);
+}
+
+bool
+LockOrStrongerHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
+{
+   return LockHeldByMeExtended(locktag, lockmode, true);
 }
 
 #ifdef USE_ASSERT_CHECKING

diff --git a/src/include/storage/lmgr.h b/src/include/storage/lmgr.h
index ddac2d2c8415ee5f98357dabe5fbe78ab7b7eb4a..0b0e02490928d05919531e8950749c02305d5a61 100644 (file)
--- a/src/include/storage/lmgr.h
+++ b/src/include/storage/lmgr.h
@@ -47,6 +47,8 @@ extern bool ConditionalLockRelation(Relation relation, LOCKMODE lockmode);
 extern void UnlockRelation(Relation relation, LOCKMODE lockmode);
 extern bool CheckRelationLockedByMe(Relation relation, LOCKMODE lockmode,
                                    bool orstronger);
+extern bool CheckRelationOidLockedByMe(Oid relid, LOCKMODE lockmode,
+                                      bool orstronger);
 extern bool LockHasWaitersRelation(Relation relation, LOCKMODE lockmode);
 
 extern void LockRelationIdForSession(LockRelId *relid, LOCKMODE lockmode);

diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index a5286fab893bd62dacb2a1b6eaecb83f8913ebc1..5c94942d96bafd7f3df2844eab3c10d7c54b3387 100644 (file)
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -561,6 +561,7 @@ extern void LockReleaseSession(LOCKMETHODID lockmethodid);
 extern void LockReleaseCurrentOwner(LOCALLOCK **locallocks, int nlocks);
 extern void LockReassignCurrentOwner(LOCALLOCK **locallocks, int nlocks);
 extern bool LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode);
+extern bool LockOrStrongerHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode);
 #ifdef USE_ASSERT_CHECKING
 extern HTAB *GetLockMethodLocalHash(void);
 #endif