Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
During heap rebuild, lock any TOAST index until end of transaction.
authorNoah Misch <noah@leadboat.com>
Sat, 21 Mar 2020 16:38:26 +0000 (09:38 -0700)
committerNoah Misch <noah@leadboat.com>
Sat, 21 Mar 2020 16:38:35 +0000 (09:38 -0700)
swap_relation_files() calls toast_get_valid_index() to find and lock
this index, just before swapping with the rebuilt TOAST index.  The
latter function releases the lock before returning.  Potential for
mischief is low; a concurrent session can issue ALTER INDEX ... SET
(fillfactor = ...), which is not alarming.  Nonetheless, changing
pg_class.relfilenode without a lock is unconventional.  Back-patch to
9.5 (all supported versions), because another fix needs this.

Discussion: https://postgr.es/m/20191226001521.GA1772687@rfd.leadboat.com

src/backend/access/heap/tuptoaster.c
src/backend/commands/cluster.c

index 0fa3c4132b3256daecc48112fb930d51a3d296a5..e8eea1b9a3fca1433a21eb9ab9e4959d5222520f 100644 (file)
@@ -1377,8 +1377,8 @@ toast_get_valid_index(Oid toastoid, LOCKMODE lock)
    validIndexOid = RelationGetRelid(toastidxs[validIndex]);
 
    /* Close the toast relation and all its indexes */
-   toast_close_indexes(toastidxs, num_indexes, lock);
-   heap_close(toastrel, lock);
+   toast_close_indexes(toastidxs, num_indexes, NoLock);
+   heap_close(toastrel, NoLock);
 
    return validIndexOid;
 }
index 9c91f2c31f7c1007089d6170021681b25909308a..1c90f8c3382838be89ffd94e3688325019614382 100644 (file)
@@ -1614,7 +1614,7 @@ finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
 
            /* Get the associated valid index to be renamed */
            toastidx = toast_get_valid_index(newrel->rd_rel->reltoastrelid,
-                                            AccessShareLock);
+                                            NoLock);
 
            /* rename the toast table ... */
            snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u",