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 1cc6714e8cb0190e74eece1be8aab4c17eec6374..6f1910a6e0f756705fcda94ac3aa124d66799690 100644 (file)
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -1046,6 +1046,7 @@ 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 e4c792228fc1afe5da55b665f86ea5ad9cda80c5..21ed483b7fa1b1a676933187014086a3b224dad5 100644 (file)
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -4304,7 +4304,10 @@ 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 a028c9ac5644753e03b4483198a219a6c9912163..a357c0c4f0f8f451d0de3e0a17f593386ff31434 100644 (file)
--- a/src/backend/commands/tablecmds.c
+++ b/src/backend/commands/tablecmds.c
@@ -3325,8 +3325,15 @@ 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
@@ -3341,6 +3348,11 @@ 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 2da91eb492d02cae1e46af163d71bb82eb1bf171..1f87b12754143e1030ef2a6f1c25a73ab415e440 100644 (file)
--- a/src/backend/storage/lmgr/lmgr.c
+++ b/src/backend/storage/lmgr/lmgr.c
@@ -336,32 +336,26 @@ 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 ec6240fbaeed309dbddb7063df848b00d5b8778f..ba66e820d06d4ec65c4bf8d901cee4ad68daaa77 100644 (file)
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -579,11 +579,17 @@ 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;
@@ -599,7 +605,35 @@ LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
                                          &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 8ab833d1d4ac3b381d08ed2a3cce9ffb68278f98..555d7fdb347613789d12a207958ce4580e91d98c 100644 (file)
--- a/src/include/storage/lmgr.h
+++ b/src/include/storage/lmgr.h
@@ -48,6 +48,8 @@ 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 8575bea25c7640c2e0c95d52749bd258087b1d57..f67056a82b57ab13c363c847b8b4de3db95192d1 100644 (file)
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -569,6 +569,7 @@ 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