Optimize vacuuming of relations with no indexes.

If there are no indexes on a relation, items can be marked LP_UNUSED
instead of LP_DEAD when pruning. This significantly reduces WAL
volume, since we no longer need to emit one WAL record for pruning
and a second to change the LP_DEAD line pointers thus created to
LP_UNUSED.

Melanie Plageman, reviewed by Andres Freund, Peter Geoghegan, and me

Discussion: https://postgr.es/m/CAAKRu_bgvb_k0gKOXWzNKWHt560R0smrGe3E8zewKPs8fiMKkw%40mail.gmail.com

src/backend/access/heap/pruneheap.c

@@ -35,6 +35,8 @@ typedef struct
 
 	/* tuple visibility test, initialized for the relation */
 	GlobalVisState *vistest;
+	/* whether or not dead items can be set LP_UNUSED during pruning */
+	bool		mark_unused_now;
 
 	TransactionId new_prune_xid;	/* new prune hint value for page */
 	TransactionId snapshotConflictHorizon;	/* latest xid removed */
@@ -67,6 +69,7 @@ static void heap_prune_record_prunable(PruneState *prstate, TransactionId xid);
 static void heap_prune_record_redirect(PruneState *prstate,
 									   OffsetNumber offnum, OffsetNumber rdoffnum);
 static void heap_prune_record_dead(PruneState *prstate, OffsetNumber offnum);
+static void heap_prune_record_dead_or_unused(PruneState *prstate, OffsetNumber offnum);
 static void heap_prune_record_unused(PruneState *prstate, OffsetNumber offnum);
 static void page_verify_redirects(Page page);
 
@@ -148,7 +151,13 @@ heap_page_prune_opt(Relation relation, Buffer buffer)
 		{
 			PruneResult presult;
 
-			heap_page_prune(relation, buffer, vistest, &presult, NULL);
+			/*
+			 * For now, pass mark_unused_now as false regardless of whether or
+			 * not the relation has indexes, since we cannot safely determine
+			 * that during on-access pruning with the current implementation.
+			 */
+			heap_page_prune(relation, buffer, vistest, false,
+							&presult, NULL);
 
 			/*
 			 * Report the number of tuples reclaimed to pgstats.  This is
@@ -193,6 +202,9 @@ heap_page_prune_opt(Relation relation, Buffer buffer)
  * (see heap_prune_satisfies_vacuum and
  * HeapTupleSatisfiesVacuum).
  *
+ * mark_unused_now indicates whether or not dead items can be set LP_UNUSED during
+ * pruning.
+ *
  * off_loc is the offset location required by the caller to use in error
  * callback.
  *
@@ -203,6 +215,7 @@ heap_page_prune_opt(Relation relation, Buffer buffer)
 void
 heap_page_prune(Relation relation, Buffer buffer,
 				GlobalVisState *vistest,
+				bool mark_unused_now,
 				PruneResult *presult,
 				OffsetNumber *off_loc)
 {
@@ -227,6 +240,7 @@ heap_page_prune(Relation relation, Buffer buffer,
 	prstate.new_prune_xid = InvalidTransactionId;
 	prstate.rel = relation;
 	prstate.vistest = vistest;
+	prstate.mark_unused_now = mark_unused_now;
 	prstate.snapshotConflictHorizon = InvalidTransactionId;
 	prstate.nredirected = prstate.ndead = prstate.nunused = 0;
 	memset(prstate.marked, 0, sizeof(prstate.marked));
@@ -306,9 +320,9 @@ heap_page_prune(Relation relation, Buffer buffer,
 		if (off_loc)
 			*off_loc = offnum;
 
-		/* Nothing to do if slot is empty or already dead */
+		/* Nothing to do if slot is empty */
 		itemid = PageGetItemId(page, offnum);
-		if (!ItemIdIsUsed(itemid) || ItemIdIsDead(itemid))
+		if (!ItemIdIsUsed(itemid))
 			continue;
 
 		/* Process this item or chain of items */
@@ -581,7 +595,17 @@ heap_prune_chain(Buffer buffer, OffsetNumber rootoffnum,
 		 * function.)
 		 */
 		if (ItemIdIsDead(lp))
+		{
+			/*
+			 * If the caller set mark_unused_now true, we can set dead line
+			 * pointers LP_UNUSED now. We don't increment ndeleted here since
+			 * the LP was already marked dead.
+			 */
+			if (unlikely(prstate->mark_unused_now))
+				heap_prune_record_unused(prstate, offnum);
+
 			break;
+		}
 
 		Assert(ItemIdIsNormal(lp));
 		htup = (HeapTupleHeader) PageGetItem(dp, lp);
@@ -715,7 +739,7 @@ heap_prune_chain(Buffer buffer, OffsetNumber rootoffnum,
 		 * redirect the root to the correct chain member.
 		 */
 		if (i >= nchain)
-			heap_prune_record_dead(prstate, rootoffnum);
+			heap_prune_record_dead_or_unused(prstate, rootoffnum);
 		else
 			heap_prune_record_redirect(prstate, rootoffnum, chainitems[i]);
 	}
@@ -726,9 +750,9 @@ heap_prune_chain(Buffer buffer, OffsetNumber rootoffnum,
 		 * item.  This can happen if the loop in heap_page_prune caused us to
 		 * visit the dead successor of a redirect item before visiting the
 		 * redirect item.  We can clean up by setting the redirect item to
-		 * DEAD state.
+		 * DEAD state or LP_UNUSED if the caller indicated.
 		 */
-		heap_prune_record_dead(prstate, rootoffnum);
+		heap_prune_record_dead_or_unused(prstate, rootoffnum);
 	}
 
 	return ndeleted;
@@ -774,6 +798,27 @@ heap_prune_record_dead(PruneState *prstate, OffsetNumber offnum)
 	prstate->marked[offnum] = true;
 }
 
+/*
+ * Depending on whether or not the caller set mark_unused_now to true, record that a
+ * line pointer should be marked LP_DEAD or LP_UNUSED. There are other cases in
+ * which we will mark line pointers LP_UNUSED, but we will not mark line
+ * pointers LP_DEAD if mark_unused_now is true.
+ */
+static void
+heap_prune_record_dead_or_unused(PruneState *prstate, OffsetNumber offnum)
+{
+	/*
+	 * If the caller set mark_unused_now to true, we can remove dead tuples
+	 * during pruning instead of marking their line pointers dead. Set this
+	 * tuple's line pointer LP_UNUSED. We hint that this option is less
+	 * likely.
+	 */
+	if (unlikely(prstate->mark_unused_now))
+		heap_prune_record_unused(prstate, offnum);
+	else
+		heap_prune_record_dead(prstate, offnum);
+}
+
 /* Record line pointer to be marked unused */
 static void
 heap_prune_record_unused(PruneState *prstate, OffsetNumber offnum)
@@ -903,13 +948,24 @@ heap_page_prune_execute(Buffer buffer,
 #ifdef USE_ASSERT_CHECKING
 
 		/*
-		 * Only heap-only tuples can become LP_UNUSED during pruning.  They
+		 * When heap_page_prune() was called, mark_unused_now may have been
-		 * don't need to be left in place as LP_DEAD items until VACUUM gets
+		 * passed as true, which allows would-be LP_DEAD items to be made
-		 * around to doing index vacuuming.
+		 * LP_UNUSED instead. This is only possible if the relation has no
+		 * indexes. If there are any dead items, then mark_unused_now was not
+		 * true and every item being marked LP_UNUSED must refer to a
+		 * heap-only tuple.
 		 */
+		if (ndead > 0)
+		{
 			Assert(ItemIdHasStorage(lp) && ItemIdIsNormal(lp));
 			htup = (HeapTupleHeader) PageGetItem(page, lp);
 			Assert(HeapTupleHeaderIsHeapOnly(htup));
+		}
+		else
+		{
+			Assert(ItemIdIsUsed(lp));
+		}
+
 #endif
 
 		ItemIdSetUnused(lp);

src/backend/access/heap/vacuumlazy.c

@@ -1036,69 +1036,6 @@ lazy_scan_heap(LVRelState *vacrel)
 
 		Assert(!prunestate.all_visible || !prunestate.has_lpdead_items);
 
-		if (vacrel->nindexes == 0)
-		{
-			/*
-			 * Consider the need to do page-at-a-time heap vacuuming when
-			 * using the one-pass strategy now.
-			 *
-			 * The one-pass strategy will never call lazy_vacuum().  The steps
-			 * performed here can be thought of as the one-pass equivalent of
-			 * a call to lazy_vacuum().
-			 */
-			if (prunestate.has_lpdead_items)
-			{
-				Size		freespace;
-
-				lazy_vacuum_heap_page(vacrel, blkno, buf, 0, vmbuffer);
-
-				/* Forget the LP_DEAD items that we just vacuumed */
-				dead_items->num_items = 0;
-
-				/*
-				 * Now perform FSM processing for blkno, and move on to next
-				 * page.
-				 *
-				 * Our call to lazy_vacuum_heap_page() will have considered if
-				 * it's possible to set all_visible/all_frozen independently
-				 * of lazy_scan_prune().  Note that prunestate was invalidated
-				 * by lazy_vacuum_heap_page() call.
-				 */
-				freespace = PageGetHeapFreeSpace(page);
-
-				UnlockReleaseBuffer(buf);
-				RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
-
-				/*
-				 * Periodically perform FSM vacuuming to make newly-freed
-				 * space visible on upper FSM pages. FreeSpaceMapVacuumRange()
-				 * vacuums the portion of the freespace map covering heap
-				 * pages from start to end - 1. Include the block we just
-				 * vacuumed by passing it blkno + 1. Overflow isn't an issue
-				 * because MaxBlockNumber + 1 is InvalidBlockNumber which
-				 * causes FreeSpaceMapVacuumRange() to vacuum freespace map
-				 * pages covering the remainder of the relation.
-				 */
-				if (blkno - next_fsm_block_to_vacuum >= VACUUM_FSM_EVERY_PAGES)
-				{
-					FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
-											blkno + 1);
-					next_fsm_block_to_vacuum = blkno + 1;
-				}
-
-				continue;
-			}
-
-			/*
-			 * There was no call to lazy_vacuum_heap_page() because pruning
-			 * didn't encounter/create any LP_DEAD items that needed to be
-			 * vacuumed.  Prune state has not been invalidated, so proceed
-			 * with prunestate-driven visibility map and FSM steps (just like
-			 * the two-pass strategy).
-			 */
-			Assert(dead_items->num_items == 0);
-		}
-
 		/*
 		 * Handle setting visibility map bit based on information from the VM
 		 * (as of last lazy_scan_skip() call), and from prunestate
@@ -1209,39 +1146,46 @@ lazy_scan_heap(LVRelState *vacrel)
 
 		/*
 		 * Final steps for block: drop cleanup lock, record free space in the
-		 * FSM
+		 * FSM.
-		 */
+		 *
-		if (prunestate.has_lpdead_items && vacrel->do_index_vacuuming)
+		 * If we will likely do index vacuuming, wait until
-		{
+		 * lazy_vacuum_heap_rel() to save free space. This doesn't just save
-			/*
+		 * us some cycles; it also allows us to record any additional free
-			 * Wait until lazy_vacuum_heap_rel() to save free space.  This
+		 * space that lazy_vacuum_heap_page() will make available in cases
-			 * doesn't just save us some cycles; it also allows us to record
+		 * where it's possible to truncate the page's line pointer array.
-			 * any additional free space that lazy_vacuum_heap_page() will
-			 * make available in cases where it's possible to truncate the
-			 * page's line pointer array.
 		 *
 		 * Note: It's not in fact 100% certain that we really will call
-			 * lazy_vacuum_heap_rel() -- lazy_vacuum() might yet opt to skip
+		 * lazy_vacuum_heap_rel() -- lazy_vacuum() might yet opt to skip index
-			 * index vacuuming (and so must skip heap vacuuming).  This is
+		 * vacuuming (and so must skip heap vacuuming).  This is deemed okay
-			 * deemed okay because it only happens in emergencies, or when
+		 * because it only happens in emergencies, or when there is very
-			 * there is very little free space anyway. (Besides, we start
+		 * little free space anyway. (Besides, we start recording free space
-			 * recording free space in the FSM once index vacuuming has been
+		 * in the FSM once index vacuuming has been abandoned.)
-			 * abandoned.)
-			 *
-			 * Note: The one-pass (no indexes) case is only supposed to make
-			 * it this far when there were no LP_DEAD items during pruning.
 		 */
-			Assert(vacrel->nindexes > 0);
+		if (vacrel->nindexes == 0
-			UnlockReleaseBuffer(buf);
+			|| !vacrel->do_index_vacuuming
-		}
+			|| !prunestate.has_lpdead_items)
-		else
 		{
 			Size		freespace = PageGetHeapFreeSpace(page);
 
 			UnlockReleaseBuffer(buf);
 			RecordPageWithFreeSpace(vacrel->rel, blkno, freespace);
+
+			/*
+			 * Periodically perform FSM vacuuming to make newly-freed space
+			 * visible on upper FSM pages. This is done after vacuuming if the
+			 * table has indexes.
+			 */
+			if (vacrel->nindexes == 0 && prunestate.has_lpdead_items &&
+				blkno - next_fsm_block_to_vacuum >= VACUUM_FSM_EVERY_PAGES)
+			{
+				FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
+										blkno);
+				next_fsm_block_to_vacuum = blkno;
 			}
 		}
+		else
+			UnlockReleaseBuffer(buf);
+	}
 
 	vacrel->blkno = InvalidBlockNumber;
 	if (BufferIsValid(vmbuffer))
@@ -1596,8 +1540,13 @@ lazy_scan_prune(LVRelState *vacrel,
 	 * in presult.ndeleted. It should not be confused with lpdead_items;
 	 * lpdead_items's final value can be thought of as the number of tuples
 	 * that were deleted from indexes.
+	 *
+	 * If the relation has no indexes, we can immediately mark would-be dead
+	 * items LP_UNUSED, so mark_unused_now should be true if no indexes and
+	 * false otherwise.
 	 */
-	heap_page_prune(rel, buf, vacrel->vistest, &presult, &vacrel->offnum);
+	heap_page_prune(rel, buf, vacrel->vistest, vacrel->nindexes == 0,
+					&presult, &vacrel->offnum);
 
 	/*
 	 * Now scan the page to collect LP_DEAD items and check for tuples
@@ -2520,7 +2469,7 @@ lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno, Buffer buffer,
 	bool		all_frozen;
 	LVSavedErrInfo saved_err_info;
 
-	Assert(vacrel->nindexes == 0 || vacrel->do_index_vacuuming);
+	Assert(vacrel->do_index_vacuuming);
 
 	pgstat_progress_update_param(PROGRESS_VACUUM_HEAP_BLKS_VACUUMED, blkno);
 

src/include/access/heapam.h

@@ -320,6 +320,7 @@ struct GlobalVisState;
 extern void heap_page_prune_opt(Relation relation, Buffer buffer);
 extern void heap_page_prune(Relation relation, Buffer buffer,
 							struct GlobalVisState *vistest,
+							bool mark_unused_now,
 							PruneResult *presult,
 							OffsetNumber *off_loc);
 extern void heap_page_prune_execute(Buffer buffer,