Fix logical decoding error when system table w/ toast is repeatedly rewritten.

Repeatedly rewriting a mapped catalog table with VACUUM FULL or
CLUSTER could cause logical decoding to fail with:
ERROR, "could not map filenode \"%s\" to relation OID"

To trigger the problem the rewritten catalog had to have live tuples
with toasted columns.

The problem was triggered as during catalog table rewrites the
heap_insert() check that prevents logical decoding information to be
emitted for system catalogs, failed to treat the new heap's toast table
as a system catalog (because the new heap is not recognized as a
catalog table via RelationIsLogicallyLogged()). The relmapper, in
contrast to the normal catalog contents, does not contain historical
information. After a single rewrite of a mapped table the new relation
is known to the relmapper, but if the table is rewritten twice before
logical decoding occurs, the relfilenode cannot be mapped to a
relation anymore.  Which then leads us to error out.   This only
happens for toast tables, because the main table contents aren't
re-inserted with heap_insert().

The fix is simple, add a new heap_insert() flag that prevents logical
decoding information from being emitted, and accept during decoding
that there might not be tuple data for toast tables.

Unfortunately that does not fix pre-existing logical decoding
errors. Doing so would require not throwing an error when a filenode
cannot be mapped to a relation during decoding, and that seems too
likely to hide bugs.  If it's crucial to fix decoding for an existing
slot, temporarily changing the ERROR in ReorderBufferCommit() to a
WARNING appears to be the best fix.

Author: Andres Freund
Discussion: https://postgr.es/m/20180914021046.oi7dm4ra3ot2g2kt@alap3.anarazel.de
Backpatch: 9.4-, where logical decoding was introduced

3 files changed, 46 insertions, 9 deletions

diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.c
index 83a03f3d0af..ca6525154ba 100644
--- a/src/backend/access/heap/heapam.c
+++ b/src/backend/access/heap/heapam.c
@@ -2130,6 +2130,11 @@ FreeBulkInsertState(BulkInsertState bistate)
  * Speculatively inserted tuples behave as "value locks" of short duration,
  * used to implement INSERT .. ON CONFLICT.
  *
+ * HEAP_INSERT_NO_LOGICAL force-disables the emitting of logical decoding
+ * information for the tuple. This should solely be used during table rewrites
+ * where RelationIsLogicallyLogged(relation) is not yet accurate for the new
+ * relation.
+ *
  * Note that most of these options will be applied when inserting into the
  * heap's TOAST table, too, if the tuple requires any out-of-line data.  Only
  * HEAP_INSERT_IS_SPECULATIVE is explicitly ignored, as the toast data does
@@ -2258,7 +2263,8 @@ heap_insert(Relation relation, HeapTuple tup, CommandId cid,
 		 * page write, so make sure it's included even if we take a full-page
 		 * image. (XXX We could alternatively store a pointer into the FPW).
 		 */
-		if (RelationIsLogicallyLogged(relation))
+		if (RelationIsLogicallyLogged(relation) &&
+			!(options & HEAP_INSERT_NO_LOGICAL))
 		{
 			xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
 			bufflags |= REGBUF_KEEP_DATA;
@@ -2421,6 +2427,9 @@ heap_multi_insert(Relation relation, HeapTuple *tuples, int ntuples,
 	bool		need_tuple_data = RelationIsLogicallyLogged(relation);
 	bool		need_cids = RelationIsAccessibleInLogicalDecoding(relation);
 
+	/* currently not needed (thus unsupported) for heap_multi_insert() */
+	AssertArg(!(options & HEAP_INSERT_NO_LOGICAL));
+
 	needwal = !(options & HEAP_INSERT_SKIP_WAL) && RelationNeedsWAL(relation);
 	saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
 												   HEAP_DEFAULT_FILLFACTOR);
diff --git a/src/backend/access/heap/rewriteheap.c b/src/backend/access/heap/rewriteheap.c
index c8fa62daeb9..ca18d986744 100644
--- a/src/backend/access/heap/rewriteheap.c
+++ b/src/backend/access/heap/rewriteheap.c
@@ -651,10 +651,23 @@ raw_heap_insert(RewriteState state, HeapTuple tup)
 		heaptup = tup;
 	}
 	else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
+	{
+		int options = HEAP_INSERT_SKIP_FSM;
+
+		if (!state->rs_use_wal)
+			options |= HEAP_INSERT_SKIP_WAL;
+
+		/*
+		 * The new relfilenode's relcache entrye doesn't have the necessary
+		 * information to determine whether a relation should emit data for
+		 * logical decoding.  Force it to off if necessary.
+		 */
+		if (!RelationIsLogicallyLogged(state->rs_old_rel))
+			options |= HEAP_INSERT_NO_LOGICAL;
+
 		heaptup = toast_insert_or_update(state->rs_new_rel, tup, NULL,
-										 HEAP_INSERT_SKIP_FSM |
-										 (state->rs_use_wal ?
-										  0 : HEAP_INSERT_SKIP_WAL));
+										 options);
+	}
 	else
 		heaptup = tup;
 
diff --git a/src/backend/replication/logical/reorderbuffer.c b/src/backend/replication/logical/reorderbuffer.c
index 982cddc006a..44d9642fd86 100644
--- a/src/backend/replication/logical/reorderbuffer.c
+++ b/src/backend/replication/logical/reorderbuffer.c
@@ -1528,8 +1528,16 @@ ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
 											change->data.tp.relnode.relNode);
 
 					/*
-					 * Catalog tuple without data, emitted while catalog was
-					 * in the process of being rewritten.
+					 * Mapped catalog tuple without data, emitted while
+					 * catalog table was in the process of being rewritten. We
+					 * can fail to look up the relfilenode, because the the
+					 * relmapper has no "historic" view, in contrast to normal
+					 * the normal catalog during decoding. Thus repeated
+					 * rewrites can cause a lookup failure. That's OK because
+					 * we do not decode catalog changes anyway. Normally such
+					 * tuples would be skipped over below, but we can't
+					 * identify whether the table should be logically logged
+					 * without mapping the relfilenode to the oid.
 					 */
 					if (reloid == InvalidOid &&
 						change->data.tp.newtuple == NULL &&
@@ -1584,10 +1592,17 @@ ReorderBufferCommit(ReorderBuffer *rb, TransactionId xid,
 						 * transaction's changes. Otherwise it will get
 						 * freed/reused while restoring spooled data from
 						 * disk.
+						 *
+						 * But skip doing so if there's no tuple-data. That
+						 * happens if a non-mapped system catalog with a toast
+						 * table is rewritten.
 						 */
-						dlist_delete(&change->node);
-						ReorderBufferToastAppendChunk(rb, txn, relation,
-													  change);
+						if (change->data.tp.newtuple != NULL)
+						{
+							dlist_delete(&change->node);
+							ReorderBufferToastAppendChunk(rb, txn, relation,
+														  change);
+						}
 					}
 
 			change_done: