Set cutoff xmin more aggressively when vacuuming a temporary table.

Since other sessions aren't allowed to look into a temporary table
of our own session, we do not need to worry about the global xmin
horizon when setting the vacuum XID cutoff.  Indeed, if we're not
inside a transaction block, we may set oldestXmin to be the next
XID, because there cannot be any in-doubt tuples in a temp table,
nor any tuples that are dead but still visible to some snapshot of
our transaction.  (VACUUM, of course, is never inside a transaction
block; but we need to test that because CLUSTER shares the same code.)

This approach allows us to always clean out a temp table completely
during VACUUM, independently of concurrent activity.  Aside from
being useful in its own right, that simplifies building reproducible
test cases.

Discussion: https://postgr.es/m/3490536.1598629609@sss.pgh.pa.us

src/backend/access/heap/vacuumlazy.c

@@ -471,6 +471,7 @@ heap_vacuum_rel(Relation onerel, VacuumParams *params,
 						  params->freeze_table_age,
 						  params->multixact_freeze_min_age,
 						  params->multixact_freeze_table_age,
+						  true, /* we must be a top-level command */
 						  &OldestXmin, &FreezeLimit, &xidFullScanLimit,
 						  &MultiXactCutoff, &mxactFullScanLimit);
 

src/backend/commands/cluster.c

@@ -67,10 +67,13 @@ typedef struct
 } RelToCluster;
 
 
-static void rebuild_relation(Relation OldHeap, Oid indexOid, bool verbose);
+static void rebuild_relation(Relation OldHeap, Oid indexOid,
+							 bool isTopLevel, bool verbose);
 static void copy_table_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex,
-							bool verbose, bool *pSwapToastByContent,
+							bool isTopLevel, bool verbose,
-							TransactionId *pFreezeXid, MultiXactId *pCutoffMulti);
+							bool *pSwapToastByContent,
+							TransactionId *pFreezeXid,
+							MultiXactId *pCutoffMulti);
 static List *get_tables_to_cluster(MemoryContext cluster_context);
 
 
@@ -170,7 +173,7 @@ cluster(ClusterStmt *stmt, bool isTopLevel)
 		table_close(rel, NoLock);
 
 		/* Do the job. */
-		cluster_rel(tableOid, indexOid, stmt->options);
+		cluster_rel(tableOid, indexOid, stmt->options, isTopLevel);
 	}
 	else
 	{
@@ -219,7 +222,8 @@ cluster(ClusterStmt *stmt, bool isTopLevel)
 			PushActiveSnapshot(GetTransactionSnapshot());
 			/* Do the job. */
 			cluster_rel(rvtc->tableOid, rvtc->indexOid,
-						stmt->options | CLUOPT_RECHECK);
+						stmt->options | CLUOPT_RECHECK,
+						isTopLevel);
 			PopActiveSnapshot();
 			CommitTransactionCommand();
 		}
@@ -250,7 +254,7 @@ cluster(ClusterStmt *stmt, bool isTopLevel)
  * and error messages should refer to the operation as VACUUM not CLUSTER.
  */
 void
-cluster_rel(Oid tableOid, Oid indexOid, int options)
+cluster_rel(Oid tableOid, Oid indexOid, int options, bool isTopLevel)
 {
 	Relation	OldHeap;
 	bool		verbose = ((options & CLUOPT_VERBOSE) != 0);
@@ -400,7 +404,7 @@ cluster_rel(Oid tableOid, Oid indexOid, int options)
 	TransferPredicateLocksToHeapRelation(OldHeap);
 
 	/* rebuild_relation does all the dirty work */
-	rebuild_relation(OldHeap, indexOid, verbose);
+	rebuild_relation(OldHeap, indexOid, isTopLevel, verbose);
 
 	/* NB: rebuild_relation does table_close() on OldHeap */
 
@@ -545,11 +549,12 @@ mark_index_clustered(Relation rel, Oid indexOid, bool is_internal)
  *
  * OldHeap: table to rebuild --- must be opened and exclusive-locked!
  * indexOid: index to cluster by, or InvalidOid to rewrite in physical order.
+ * isTopLevel: should be passed down from ProcessUtility.
  *
  * NB: this routine closes OldHeap at the right time; caller should not.
  */
 static void
-rebuild_relation(Relation OldHeap, Oid indexOid, bool verbose)
+rebuild_relation(Relation OldHeap, Oid indexOid, bool isTopLevel, bool verbose)
 {
 	Oid			tableOid = RelationGetRelid(OldHeap);
 	Oid			tableSpace = OldHeap->rd_rel->reltablespace;
@@ -577,7 +582,7 @@ rebuild_relation(Relation OldHeap, Oid indexOid, bool verbose)
 							   AccessExclusiveLock);
 
 	/* Copy the heap data into the new table in the desired order */
-	copy_table_data(OIDNewHeap, tableOid, indexOid, verbose,
+	copy_table_data(OIDNewHeap, tableOid, indexOid, isTopLevel, verbose,
 					&swap_toast_by_content, &frozenXid, &cutoffMulti);
 
 	/*
@@ -728,7 +733,8 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, char relpersistence,
  * *pCutoffMulti receives the MultiXactId used as a cutoff point.
  */
 static void
-copy_table_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex, bool verbose,
+copy_table_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex,
+				bool isTopLevel, bool verbose,
 				bool *pSwapToastByContent, TransactionId *pFreezeXid,
 				MultiXactId *pCutoffMulti)
 {
@@ -826,7 +832,7 @@ copy_table_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex, bool verbose,
 	 * Since we're going to rewrite the whole table anyway, there's no reason
 	 * not to be aggressive about this.
 	 */
-	vacuum_set_xid_limits(OldHeap, 0, 0, 0, 0,
+	vacuum_set_xid_limits(OldHeap, 0, 0, 0, 0, isTopLevel,
 						  &OldestXmin, &FreezeXid, NULL, &MultiXactCutoff,
 						  NULL);
 

src/backend/commands/vacuum.c

@@ -907,6 +907,9 @@ get_all_vacuum_rels(int options)
 /*
  * vacuum_set_xid_limits() -- compute oldestXmin and freeze cutoff points
  *
+ * Input parameters are the target relation, applicable freeze age settings,
+ * and isTopLevel which should be passed down from ProcessUtility.
+ *
  * The output parameters are:
  * - oldestXmin is the cutoff value used to distinguish whether tuples are
  *	 DEAD or RECENTLY_DEAD (see HeapTupleSatisfiesVacuum).
@@ -931,6 +934,7 @@ vacuum_set_xid_limits(Relation rel,
 					  int freeze_table_age,
 					  int multixact_freeze_min_age,
 					  int multixact_freeze_table_age,
+					  bool isTopLevel,
 					  TransactionId *oldestXmin,
 					  TransactionId *freezeLimit,
 					  TransactionId *xidFullScanLimit,
@@ -946,14 +950,32 @@ vacuum_set_xid_limits(Relation rel,
 	MultiXactId mxactLimit;
 	MultiXactId safeMxactLimit;
 
+	if (RELATION_IS_LOCAL(rel) && !IsInTransactionBlock(isTopLevel))
+	{
 		/*
-	 * We can always ignore processes running lazy vacuum.  This is because we
+		 * If we are processing a temp relation (which by prior checks must be
-	 * use these values only for deciding which tuples we must keep in the
+		 * one belonging to our session), and we are not inside any
-	 * tables.  Since lazy vacuum doesn't write its XID anywhere (usually no
+		 * transaction block, then there can be no tuples in the rel that are
-	 * XID assigned), it's safe to ignore it.  In theory it could be
+		 * still in-doubt, nor can there be any that are dead but possibly
-	 * problematic to ignore lazy vacuums in a full vacuum, but keep in mind
+		 * still interesting to some snapshot our session holds.  We don't
-	 * that only one vacuum process can be working on a particular table at
+		 * need to care whether other sessions could see such tuples, either.
-	 * any time, and that each vacuum is always an independent transaction.
+		 * So we can aggressively set the cutoff xmin to be the nextXid.
+		 */
+		*oldestXmin = ReadNewTransactionId();
+	}
+	else
+	{
+		/*
+		 * Otherwise, calculate the cutoff xmin normally.
+		 *
+		 * We can always ignore processes running lazy vacuum.  This is
+		 * because we use these values only for deciding which tuples we must
+		 * keep in the tables.  Since lazy vacuum doesn't write its XID
+		 * anywhere (usually no XID assigned), it's safe to ignore it.  In
+		 * theory it could be problematic to ignore lazy vacuums in a full
+		 * vacuum, but keep in mind that only one vacuum process can be
+		 * working on a particular table at any time, and that each vacuum is
+		 * always an independent transaction.
 		 */
 		*oldestXmin = GetOldestNonRemovableTransactionId(rel);
 
@@ -962,18 +984,21 @@ vacuum_set_xid_limits(Relation rel,
 			TransactionId limit_xmin;
 			TimestampTz limit_ts;
 
-		if (TransactionIdLimitedForOldSnapshots(*oldestXmin, rel, &limit_xmin, &limit_ts))
+			if (TransactionIdLimitedForOldSnapshots(*oldestXmin, rel,
+													&limit_xmin, &limit_ts))
 			{
 				/*
-			 * TODO: We should only set the threshold if we are pruning on the
+				 * TODO: We should only set the threshold if we are pruning on
-			 * basis of the increased limits. Not as crucial here as it is for
+				 * the basis of the increased limits.  Not as crucial here as
-			 * opportunistic pruning (which often happens at a much higher
+				 * it is for opportunistic pruning (which often happens at a
-			 * frequency), but would still be a significant improvement.
+				 * much higher frequency), but would still be a significant
+				 * improvement.
 				 */
 				SetOldSnapshotThresholdTimestamp(limit_ts, limit_xmin);
 				*oldestXmin = limit_xmin;
 			}
 		}
+	}
 
 	Assert(TransactionIdIsNormal(*oldestXmin));
 
@@ -1905,7 +1930,7 @@ vacuum_rel(Oid relid, RangeVar *relation, VacuumParams *params)
 			cluster_options |= CLUOPT_VERBOSE;
 
 		/* VACUUM FULL is now a variant of CLUSTER; see cluster.c */
-		cluster_rel(relid, InvalidOid, cluster_options);
+		cluster_rel(relid, InvalidOid, cluster_options, true);
 	}
 	else
 		table_relation_vacuum(onerel, params, vac_strategy);

src/include/commands/cluster.h

@@ -19,7 +19,8 @@
 
 
 extern void cluster(ClusterStmt *stmt, bool isTopLevel);
-extern void cluster_rel(Oid tableOid, Oid indexOid, int options);
+extern void cluster_rel(Oid tableOid, Oid indexOid, int options,
+						bool isTopLevel);
 extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid,
 									   bool recheck, LOCKMODE lockmode);
 extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal);

src/include/commands/vacuum.h

@@ -267,6 +267,7 @@ extern void vacuum_set_xid_limits(Relation rel,
 								  int freeze_min_age, int freeze_table_age,
 								  int multixact_freeze_min_age,
 								  int multixact_freeze_table_age,
+								  bool isTopLevel,
 								  TransactionId *oldestXmin,
 								  TransactionId *freezeLimit,
 								  TransactionId *xidFullScanLimit,