Use an MVCC snapshot, rather than SnapshotNow, for catalog scans.

SnapshotNow scans have the undesirable property that, in the face of
concurrent updates, the scan can fail to see either the old or the new
versions of the row.  In many cases, we work around this by requiring
DDL operations to hold AccessExclusiveLock on the object being
modified; in some cases, the existing locking is inadequate and random
failures occur as a result.  This commit doesn't change anything
related to locking, but will hopefully pave the way to allowing lock
strength reductions in the future.

The major issue has held us back from making this change in the past
is that taking an MVCC snapshot is significantly more expensive than
using a static special snapshot such as SnapshotNow.  However, testing
of various worst-case scenarios reveals that this problem is not
severe except under fairly extreme workloads.  To mitigate those
problems, we avoid retaking the MVCC snapshot for each new scan;
instead, we take a new snapshot only when invalidation messages have
been processed.  The catcache machinery already requires that
invalidation messages be sent before releasing the related heavyweight
lock; else other backends might rely on locally-cached data rather
than scanning the catalog at all.  Thus, making snapshot reuse
dependent on the same guarantees shouldn't break anything that wasn't
already subtly broken.

Patch by me.  Review by Michael Paquier and Andres Freund.

src/backend/access/heap/heapam.c

@@ -80,7 +80,7 @@ static HeapScanDesc heap_beginscan_internal(Relation relation,
 						Snapshot snapshot,
 						int nkeys, ScanKey key,
 						bool allow_strat, bool allow_sync,
-						bool is_bitmapscan);
+						bool is_bitmapscan, bool temp_snap);
 static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
 					TransactionId xid, CommandId cid, int options);
 static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
@@ -1286,7 +1286,17 @@ heap_beginscan(Relation relation, Snapshot snapshot,
 			   int nkeys, ScanKey key)
 {
 	return heap_beginscan_internal(relation, snapshot, nkeys, key,
-								   true, true, false);
+								   true, true, false, false);
+}
+
+HeapScanDesc
+heap_beginscan_catalog(Relation relation, int nkeys, ScanKey key)
+{
+	Oid			relid = RelationGetRelid(relation);
+	Snapshot	snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
+
+	return heap_beginscan_internal(relation, snapshot, nkeys, key,
+								   true, true, false, true);
 }
 
 HeapScanDesc
@@ -1295,7 +1305,7 @@ heap_beginscan_strat(Relation relation, Snapshot snapshot,
 					 bool allow_strat, bool allow_sync)
 {
 	return heap_beginscan_internal(relation, snapshot, nkeys, key,
-								   allow_strat, allow_sync, false);
+								   allow_strat, allow_sync, false, false);
 }
 
 HeapScanDesc
@@ -1303,14 +1313,14 @@ heap_beginscan_bm(Relation relation, Snapshot snapshot,
 				  int nkeys, ScanKey key)
 {
 	return heap_beginscan_internal(relation, snapshot, nkeys, key,
-								   false, false, true);
+								   false, false, true, false);
 }
 
 static HeapScanDesc
 heap_beginscan_internal(Relation relation, Snapshot snapshot,
 						int nkeys, ScanKey key,
 						bool allow_strat, bool allow_sync,
-						bool is_bitmapscan)
+						bool is_bitmapscan, bool temp_snap)
 {
 	HeapScanDesc scan;
 
@@ -1335,6 +1345,7 @@ heap_beginscan_internal(Relation relation, Snapshot snapshot,
 	scan->rs_strategy = NULL;	/* set in initscan */
 	scan->rs_allow_strat = allow_strat;
 	scan->rs_allow_sync = allow_sync;
+	scan->rs_temp_snap = temp_snap;
 
 	/*
 	 * we can use page-at-a-time mode if it's an MVCC-safe snapshot
@@ -1421,6 +1432,9 @@ heap_endscan(HeapScanDesc scan)
 	if (scan->rs_strategy != NULL)
 		FreeAccessStrategy(scan->rs_strategy);
 
+	if (scan->rs_temp_snap)
+		UnregisterSnapshot(scan->rs_snapshot);
+
 	pfree(scan);
 }
 

src/backend/access/index/genam.c

@@ -28,6 +28,7 @@
 #include "utils/builtins.h"
 #include "utils/lsyscache.h"
 #include "utils/rel.h"
+#include "utils/snapmgr.h"
 #include "utils/tqual.h"
 
 
@@ -231,7 +232,7 @@ BuildIndexValueDescription(Relation indexRelation,
  *	rel: catalog to scan, already opened and suitably locked
  *	indexId: OID of index to conditionally use
  *	indexOK: if false, forces a heap scan (see notes below)
- *	snapshot: time qual to use (usually should be SnapshotNow)
+ *	snapshot: time qual to use (NULL for a recent catalog snapshot)
  *	nkeys, key: scan keys
  *
  * The attribute numbers in the scan key should be set for the heap case.
@@ -266,6 +267,19 @@ systable_beginscan(Relation heapRelation,
 	sysscan->heap_rel = heapRelation;
 	sysscan->irel = irel;
 
+	if (snapshot == NULL)
+	{
+		Oid		relid = RelationGetRelid(heapRelation);
+
+		snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
+		sysscan->snapshot = snapshot;
+	}
+	else
+	{
+		/* Caller is responsible for any snapshot. */
+		sysscan->snapshot = NULL;
+	}
+
 	if (irel)
 	{
 		int			i;
@@ -401,6 +415,9 @@ systable_endscan(SysScanDesc sysscan)
 	else
 		heap_endscan(sysscan->scan);
 
+	if (sysscan->snapshot)
+		UnregisterSnapshot(sysscan->snapshot);
+
 	pfree(sysscan);
 }
 
@@ -444,6 +461,19 @@ systable_beginscan_ordered(Relation heapRelation,
 	sysscan->heap_rel = heapRelation;
 	sysscan->irel = indexRelation;
 
+	if (snapshot == NULL)
+	{
+		Oid		relid = RelationGetRelid(heapRelation);
+
+		snapshot = RegisterSnapshot(GetCatalogSnapshot(relid));
+		sysscan->snapshot = snapshot;
+	}
+	else
+	{
+		/* Caller is responsible for any snapshot. */
+		sysscan->snapshot = NULL;
+	}
+
 	/* Change attribute numbers to be index column numbers. */
 	for (i = 0; i < nkeys; i++)
 	{
@@ -494,5 +524,7 @@ systable_endscan_ordered(SysScanDesc sysscan)
 {
 	Assert(sysscan->irel);
 	index_endscan(sysscan->iscan);
+	if (sysscan->snapshot)
+		UnregisterSnapshot(sysscan->snapshot);
 	pfree(sysscan);
 }

src/backend/access/nbtree/README

@@ -141,9 +141,10 @@ deletes index entries before deleting tuples, the super-exclusive lock
 guarantees that VACUUM can't delete any heap tuple that an indexscanning
 process might be about to visit.  (This guarantee works only for simple
 indexscans that visit the heap in sync with the index scan, not for bitmap
-scans.  We only need the guarantee when using non-MVCC snapshot rules such
-as SnapshotNow, so in practice this is only important for system catalog
-accesses.)
+scans.  We only need the guarantee when using non-MVCC snapshot rules; in
+an MVCC snapshot, it wouldn't matter if the heap tuple were replaced with
+an unrelated tuple at the same TID, because the new tuple wouldn't be
+visible to our scan anyway.)
 
 Because a page can be split even while someone holds a pin on it, it is
 possible that an indexscan will return items that are no longer stored on

src/backend/access/rmgrdesc/xactdesc.c

@@ -69,11 +69,14 @@ xact_desc_commit(StringInfo buf, xl_xact_commit *xlrec)
 				appendStringInfo(buf, " catalog %u", msg->cat.catId);
 			else if (msg->id == SHAREDINVALRELCACHE_ID)
 				appendStringInfo(buf, " relcache %u", msg->rc.relId);
-			/* remaining cases not expected, but print something anyway */
+			/* not expected, but print something anyway */
 			else if (msg->id == SHAREDINVALSMGR_ID)
 				appendStringInfo(buf, " smgr");
+			/* not expected, but print something anyway */
 			else if (msg->id == SHAREDINVALRELMAP_ID)
 				appendStringInfo(buf, " relmap");
+			else if (msg->id == SHAREDINVALSNAPSHOT_ID)
+				appendStringInfo(buf, " snapshot %u", msg->sn.relId);
 			else
 				appendStringInfo(buf, " unknown id %d", msg->id);
 		}