nbtree VACUUM: Cope with buggy opclasses.

petergeoghegan · petergeoghegan · commit 5b861baa550a · 2021-03-23T16:09:51.000-07:00
Teach nbtree VACUUM to press on with vacuuming in the event of a page deletion attempt that fails to "re-find" a downlink for its child/target page. There is no good reason to treat this as an irrecoverable error. But there is a good reason not to: pressing on at this point removes any question of VACUUM not making progress solely due to misbehavior from user-defined operator class code. Discussion: https://postgr.es/m/CAH2-Wzma5G9CTtMjbrXTwOym+U=aWg-R7=-htySuztgoJLvZXg@mail.gmail.com
diff --git a/src/backend/access/nbtree/nbtpage.c b/src/backend/access/nbtree/nbtpage.c
@@ -2791,10 +2791,26 @@ _bt_lock_subtree_parent(Relation rel, BlockNumber child, BTStack stack,
 	 */
 	pbuf = _bt_getstackbuf(rel, stack, child);
 	if (pbuf == InvalidBuffer)
-		ereport(ERROR,
+	{
+		/*
+		 * Failed to "re-find" a pivot tuple whose downlink matched our child
+		 * block number on the parent level -- the index must be corrupt.
+		 * Don't even try to delete the leafbuf subtree.  Just report the
+		 * issue and press on with vacuuming the index.
+		 *
+		 * Note: _bt_getstackbuf() recovers from concurrent page splits that
+		 * take place on the parent level.  Its approach is a near-exhaustive
+		 * linear search.  This also gives it a surprisingly good chance of
+		 * recovering in the event of a buggy or inconsistent opclass.  But we
+		 * don't rely on that here.
+		 */
+		ereport(LOG,
 				(errcode(ERRCODE_INDEX_CORRUPTED),
 				 errmsg_internal("failed to re-find parent key in index \"%s\" for deletion target page %u",
 								 RelationGetRelationName(rel), child)));
+		return false;
+	}
+
 	parent = stack->bts_blkno;
 	parentoffset = stack->bts_offset;
 
