Allow REPLICA IDENTITY to be set on an index that's not (yet) valid.

The motivation for this change is that when pg_dump dumps a
partitioned index that's marked REPLICA IDENTITY, it generates a
command sequence that applies REPLICA IDENTITY before the partitioned
index has been marked valid, causing restore to fail.  We could
perhaps change pg_dump to not do it like that, but that would be
difficult and would not fix existing dump files with the problem.
There seems to be very little reason for the backend to disallow
this anyway --- the code ignores indisreplident when the index
isn't valid --- so instead let's fix it by allowing the case.

Commit 9511fb37a previously expressed a concern that allowing
indisreplident to be set on invalid indexes might allow us to
wind up in a situation where a table could have indisreplident
set on multiple indexes.  I'm not sure I follow that concern
exactly, but in any case the only way that could happen is because
relation_mark_replica_identity is too trusting about the existing set
of markings being valid.  Let's just rip out its early-exit code path
(which sure looks like premature optimization anyway; what are we
doing expending code to make redundant ALTER TABLE ... REPLICA
IDENTITY commands marginally faster and not-redundant ones marginally
slower?) and fix it to positively guarantee that no more than one
index is marked indisreplident.

The pg_dump failure can be demonstrated in all supported branches,
so back-patch all the way.  I chose to back-patch 9511fb37a as well,
just to keep indisreplident handling the same in all branches.

Per bug #17756 from Sergey Belyashov.

Discussion: https://postgr.es/m/17756-dd50e8e0c8dd4a40@postgresql.org

1 files changed, 22 insertions, 42 deletions

diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
index 7c697a285bd..1293545947e 100644
--- a/src/backend/commands/tablecmds.c
+++ b/src/backend/commands/tablecmds.c
@@ -15773,7 +15773,10 @@ ATExecDropOf(Relation rel, LOCKMODE lockmode)
  * relation_mark_replica_identity: Update a table's replica identity
  *
  * Iff ri_type = REPLICA_IDENTITY_INDEX, indexOid must be the Oid of a suitable
- * index. Otherwise, it should be InvalidOid.
+ * index. Otherwise, it must be InvalidOid.
+ *
+ * Caller had better hold an exclusive lock on the relation, as the results
+ * of running two of these concurrently wouldn't be pretty.
  */
 static void
 relation_mark_replica_identity(Relation rel, char ri_type, Oid indexOid,
@@ -15785,7 +15788,6 @@ relation_mark_replica_identity(Relation rel, char ri_type, Oid indexOid,
 	HeapTuple	pg_index_tuple;
 	Form_pg_class pg_class_form;
 	Form_pg_index pg_index_form;
-
 	ListCell   *index;
 
 	/*
@@ -15807,29 +15809,7 @@ relation_mark_replica_identity(Relation rel, char ri_type, Oid indexOid,
 	heap_freetuple(pg_class_tuple);
 
 	/*
-	 * Check whether the correct index is marked indisreplident; if so, we're
-	 * done.
-	 */
-	if (OidIsValid(indexOid))
-	{
-		Assert(ri_type == REPLICA_IDENTITY_INDEX);
-
-		pg_index_tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(indexOid));
-		if (!HeapTupleIsValid(pg_index_tuple))
-			elog(ERROR, "cache lookup failed for index %u", indexOid);
-		pg_index_form = (Form_pg_index) GETSTRUCT(pg_index_tuple);
-
-		if (pg_index_form->indisreplident)
-		{
-			ReleaseSysCache(pg_index_tuple);
-			return;
-		}
-		ReleaseSysCache(pg_index_tuple);
-	}
-
-	/*
-	 * Clear the indisreplident flag from any index that had it previously,
-	 * and set it for any index that should have it now.
+	 * Update the per-index indisreplident flags correctly.
 	 */
 	pg_index = table_open(IndexRelationId, RowExclusiveLock);
 	foreach(index, RelationGetIndexList(rel))
@@ -15843,19 +15823,23 @@ relation_mark_replica_identity(Relation rel, char ri_type, Oid indexOid,
 			elog(ERROR, "cache lookup failed for index %u", thisIndexOid);
 		pg_index_form = (Form_pg_index) GETSTRUCT(pg_index_tuple);
 
-		/*
-		 * Unset the bit if set.  We know it's wrong because we checked this
-		 * earlier.
-		 */
-		if (pg_index_form->indisreplident)
+		if (thisIndexOid == indexOid)
 		{
-			dirty = true;
-			pg_index_form->indisreplident = false;
+			/* Set the bit if not already set. */
+			if (!pg_index_form->indisreplident)
+			{
+				dirty = true;
+				pg_index_form->indisreplident = true;
+			}
 		}
-		else if (thisIndexOid == indexOid)
+		else
 		{
-			dirty = true;
-			pg_index_form->indisreplident = true;
+			/* Unset the bit if set. */
+			if (pg_index_form->indisreplident)
+			{
+				dirty = true;
+				pg_index_form->indisreplident = false;
+			}
 		}
 
 		if (dirty)
@@ -15867,7 +15851,9 @@ relation_mark_replica_identity(Relation rel, char ri_type, Oid indexOid,
 			/*
 			 * Invalidate the relcache for the table, so that after we commit
 			 * all sessions will refresh the table's replica identity index
-			 * before attempting any UPDATE or DELETE on the table.
+			 * before attempting any UPDATE or DELETE on the table.  (If we
+			 * changed the table's pg_class row above, then a relcache inval
+			 * is already queued due to that; but we might not have.)
 			 */
 			CacheInvalidateRelcache(rel);
 		}
@@ -15952,12 +15938,6 @@ ATExecReplicaIdentity(Relation rel, ReplicaIdentityStmt *stmt, LOCKMODE lockmode
 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
 				 errmsg("cannot use partial index \"%s\" as replica identity",
 						RelationGetRelationName(indexRel))));
-	/* And neither are invalid indexes. */
-	if (!indexRel->rd_index->indisvalid)
-		ereport(ERROR,
-				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-				 errmsg("cannot use invalid index \"%s\" as replica identity",
-						RelationGetRelationName(indexRel))));
 
 	/* Check index for nullable columns. */
 	for (key = 0; key < IndexRelationGetNumberOfKeyAttributes(indexRel); key++)