Teach VACUUM to bypass unnecessary index vacuuming.

petergeoghegan · petergeoghegan · commit 5100010ee4d5 · 2021-04-07T16:14:54.000-07:00
VACUUM has never needed to call ambulkdelete() for each index in cases where there are precisely zero TIDs in its dead_tuples array by the end of its first pass over the heap (also its only pass over the heap in this scenario). Index vacuuming is simply not required when this happens. Index cleanup will still go ahead, but in practice most calls to amvacuumcleanup() are usually no-ops when there were zero preceding ambulkdelete() calls. In short, VACUUM has generally managed to avoid index scans when there were clearly no index tuples to delete from indexes. But cases with _close to_ no index tuples to delete were another matter -- a round of ambulkdelete() calls took place (one per index), each of which performed a full index scan. VACUUM now behaves just as if there were zero index tuples to delete in cases where there are in fact "virtually zero" such tuples. That is, it can now bypass index vacuuming and heap vacuuming as an optimization (though not index cleanup). Whether or not VACUUM bypasses indexes is determined dynamically, based on the just-observed number of heap pages in the table that have one or more LP_DEAD items (LP_DEAD items in heap pages have a 1:1 correspondence with index tuples that still need to be deleted from each index in the worst case). We only skip index vacuuming when 2% or less of the table's pages have one or more LP_DEAD items -- bypassing index vacuuming as an optimization must not noticeably impede setting bits in the visibility map. As a further condition, the dead_tuples array (i.e. VACUUM's array of LP_DEAD item TIDs) must not exceed 32MB at the point that the first pass over the heap finishes, which is also when the decision to bypass is made. (The VACUUM must also have been able to fit all TIDs in its maintenance_work_mem-bound dead_tuples space, though with a default maintenance_work_mem setting it can't matter.) This avoids surprising jumps in the duration and overhead of routine vacuuming with workloads where successive VACUUM operations consistently have almost zero dead index tuples. The number of LP_DEAD items may well accumulate over multiple VACUUM operations, before finally the threshold is crossed and VACUUM performs conventional index vacuuming. Even then, the optimization will have avoided a great deal of largely unnecessary index vacuuming. In the future we may teach VACUUM to skip index vacuuming on a per-index basis, using a much more sophisticated approach. For now we only consider the extreme cases, where we can be quite confident that index vacuuming just isn't worth it using simple heuristics. Also log information about how many heap pages have one or more LP_DEAD items when autovacuum logging is enabled. Author: Masahiko Sawada <sawada.mshk@gmail.com> Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAD21AoD0SkE11fMw4jD4RENAwBMcw1wasVnwpJVw3tVqPOQgAw@mail.gmail.com Discussion: https://postgr.es/m/CAH2-WzmkebqPd4MVGuPTOS9bMFvp9MDs5cRTCOsv1rQJ3jCbXw@mail.gmail.com
diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
@@ -103,6 +103,12 @@
 #define VACUUM_TRUNCATE_LOCK_WAIT_INTERVAL		50	/* ms */
 #define VACUUM_TRUNCATE_LOCK_TIMEOUT			5000	/* ms */
 
+/*
+ * Threshold that controls whether we bypass index vacuuming and heap
+ * vacuuming as an optimization
+ */
+#define BYPASS_THRESHOLD_PAGES	0.02	/* i.e. 2% of rel_pages */
+
 /*
  * When a table is small (i.e. smaller than this), save cycles by avoiding
  * repeated failsafe checks
@@ -401,7 +407,7 @@ static void lazy_scan_prune(LVRelState *vacrel, Buffer buf,
 							BlockNumber blkno, Page page,
 							GlobalVisState *vistest,
 							LVPagePruneState *prunestate);
-static void lazy_vacuum(LVRelState *vacrel);
+static void lazy_vacuum(LVRelState *vacrel, bool onecall);
 static bool lazy_vacuum_all_indexes(LVRelState *vacrel);
 static void lazy_vacuum_heap_rel(LVRelState *vacrel);
 static int	lazy_vacuum_heap_page(LVRelState *vacrel, BlockNumber blkno,
@@ -760,6 +766,31 @@ heap_vacuum_rel(Relation rel, VacuumParams *params,
 							 (long long) VacuumPageHit,
 							 (long long) VacuumPageMiss,
 							 (long long) VacuumPageDirty);
+			if (vacrel->rel_pages > 0)
+			{
+				if (vacrel->do_index_vacuuming)
+				{
+					msgfmt = _(" %u pages from table (%.2f%% of total) had %lld dead item identifiers removed\n");
+
+					if (vacrel->nindexes == 0 || vacrel->num_index_scans == 0)
+						appendStringInfo(&buf, _("index scan not needed:"));
+					else
+						appendStringInfo(&buf, _("index scan needed:"));
+				}
+				else
+				{
+					msgfmt = _(" %u pages from table (%.2f%% of total) have %lld dead item identifiers\n");
+
+					if (!vacrel->do_failsafe)
+						appendStringInfo(&buf, _("index scan bypassed:"));
+					else
+						appendStringInfo(&buf, _("index scan bypassed by failsafe:"));
+				}
+				appendStringInfo(&buf, msgfmt,
+								 vacrel->lpdead_item_pages,
+								 100.0 * vacrel->lpdead_item_pages / vacrel->rel_pages,
+								 (long long) vacrel->lpdead_items);
+			}
 			for (int i = 0; i < vacrel->nindexes; i++)
 			{
 				IndexBulkDeleteResult *istat = vacrel->indstats[i];
@@ -850,7 +881,8 @@ lazy_scan_heap(LVRelState *vacrel, VacuumParams *params, bool aggressive)
 				next_fsm_block_to_vacuum;
 	PGRUsage	ru0;
 	Buffer		vmbuffer = InvalidBuffer;
-	bool		skipping_blocks;
+	bool		skipping_blocks,
+				have_vacuumed_indexes = false;
 	StringInfoData buf;
 	const int	initprog_index[] = {
 		PROGRESS_VACUUM_PHASE,
@@ -1108,7 +1140,8 @@ lazy_scan_heap(LVRelState *vacrel, VacuumParams *params, bool aggressive)
 			}
 
 			/* Remove the collected garbage tuples from table and indexes */
-			lazy_vacuum(vacrel);
+			lazy_vacuum(vacrel, false);
+			have_vacuumed_indexes = true;
 
 			/*
 			 * Vacuum the Free Space Map to make newly-freed space visible on
@@ -1475,9 +1508,10 @@ lazy_scan_heap(LVRelState *vacrel, VacuumParams *params, bool aggressive)
 			 * Note: It's not in fact 100% certain that we really will call
 			 * lazy_vacuum_heap_rel() -- lazy_vacuum() might yet opt to skip
 			 * index vacuuming (and so must skip heap vacuuming).  This is
-			 * deemed okay because it only happens in emergencies. (Besides,
-			 * we start recording free space in the FSM once index vacuuming
-			 * has been abandoned.)
+			 * deemed okay because it only happens in emergencies, or when
+			 * there is very little free space anyway. (Besides, we start
+			 * recording free space in the FSM once index vacuuming has been
+			 * 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.
@@ -1522,9 +1556,8 @@ lazy_scan_heap(LVRelState *vacrel, VacuumParams *params, bool aggressive)
 	}
 
 	/* If any tuples need to be deleted, perform final vacuum cycle */
-	/* XXX put a threshold on min number of tuples here? */
 	if (dead_tuples->num_tuples > 0)
-		lazy_vacuum(vacrel);
+		lazy_vacuum(vacrel, !have_vacuumed_indexes);
 
 	/*
 	 * Vacuum the remainder of the Free Space Map.  We must do this whether or
@@ -1555,6 +1588,16 @@ lazy_scan_heap(LVRelState *vacrel, VacuumParams *params, bool aggressive)
 	 * If table has no indexes and at least one heap pages was vacuumed, make
 	 * log report that lazy_vacuum_heap_rel would've made had there been
 	 * indexes (having indexes implies using the two pass strategy).
+	 *
+	 * We deliberately don't do this in the case where there are indexes but
+	 * index vacuuming was bypassed.  We make a similar report at the point
+	 * that index vacuuming is bypassed, but that's actually quite different
+	 * in one important sense: it shows information about work we _haven't_
+	 * done.
+	 *
+	 * log_autovacuum output does things differently; it consistently presents
+	 * information about LP_DEAD items for the VACUUM as a whole.  We always
+	 * report on each round of index and heap vacuuming separately, though.
 	 */
 	if (vacrel->nindexes == 0 && vacrel->lpdead_item_pages > 0)
 		ereport(elevel,
@@ -1983,14 +2026,21 @@ lazy_scan_prune(LVRelState *vacrel,
 /*
  * Remove the collected garbage tuples from the table and its indexes.
  *
+ * We may choose to bypass index vacuuming at this point, though only when the
+ * ongoing VACUUM operation will definitely only have one index scan/round of
+ * index vacuuming.  Caller indicates whether or not this is such a VACUUM
+ * operation using 'onecall' argument.
+ *
  * In rare emergencies, the ongoing VACUUM operation can be made to skip both
  * index vacuuming and index cleanup at the point we're called.  This avoids
  * having the whole system refuse to allocate further XIDs/MultiXactIds due to
  * wraparound.
  */
 static void
-lazy_vacuum(LVRelState *vacrel)
+lazy_vacuum(LVRelState *vacrel, bool onecall)
 {
+	bool		do_bypass_optimization;
+
 	/* Should not end up here with no indexes */
 	Assert(vacrel->nindexes > 0);
 	Assert(!IsParallelWorker());
@@ -2003,7 +2053,75 @@ lazy_vacuum(LVRelState *vacrel)
 		return;
 	}
 
-	if (lazy_vacuum_all_indexes(vacrel))
+	/*
+	 * Consider bypassing index vacuuming (and heap vacuuming) entirely.
+	 *
+	 * We currently only do this in cases where the number of LP_DEAD items
+	 * for the entire VACUUM operation is close to zero.  This avoids sharp
+	 * discontinuities in the duration and overhead of successive VACUUM
+	 * operations that run against the same table with a fixed workload.
+	 * Ideally, successive VACUUM operations will behave as if there are
+	 * exactly zero LP_DEAD items in cases where there are close to zero.
+	 *
+	 * This is likely to be helpful with a table that is continually affected
+	 * by UPDATEs that can mostly apply the HOT optimization, but occasionally
+	 * have small aberrations that lead to just a few heap pages retaining
+	 * only one or two LP_DEAD items.  This is pretty common; even when the
+	 * DBA goes out of their way to make UPDATEs use HOT, it is practically
+	 * impossible to predict whether HOT will be applied in 100% of cases.
+	 * It's far easier to ensure that 99%+ of all UPDATEs against a table use
+	 * HOT through careful tuning.
+	 */
+	do_bypass_optimization = false;
+	if (onecall && vacrel->rel_pages > 0)
+	{
+		BlockNumber threshold;
+
+		Assert(vacrel->num_index_scans == 0);
+		Assert(vacrel->lpdead_items == vacrel->dead_tuples->num_tuples);
+		Assert(vacrel->do_index_vacuuming);
+		Assert(vacrel->do_index_cleanup);
+
+		/*
+		 * This crossover point at which we'll start to do index vacuuming is
+		 * expressed as a percentage of the total number of heap pages in the
+		 * table that are known to have at least one LP_DEAD item.  This is
+		 * much more important than the total number of LP_DEAD items, since
+		 * it's a proxy for the number of heap pages whose visibility map bits
+		 * cannot be set on account of bypassing index and heap vacuuming.
+		 *
+		 * We apply one further precautionary test: the space currently used
+		 * to store the TIDs (TIDs that now all point to LP_DEAD items) must
+		 * not exceed 32MB.  This limits the risk that we will bypass index
+		 * vacuuming again and again until eventually there is a VACUUM whose
+		 * dead_tuples space is not CPU cache resident.
+		 */
+		threshold = (double) vacrel->rel_pages * BYPASS_THRESHOLD_PAGES;
+		do_bypass_optimization =
+			(vacrel->lpdead_item_pages < threshold &&
+			 vacrel->lpdead_items < MAXDEADTUPLES(32L * 1024L * 1024L));
+	}
+
+	if (do_bypass_optimization)
+	{
+		/*
+		 * There are almost zero TIDs.  Behave as if there were precisely
+		 * zero: bypass index vacuuming, but do index cleanup.
+		 *
+		 * We expect that the ongoing VACUUM operation will finish very
+		 * quickly, so there is no point in considering speeding up as a
+		 * failsafe against wraparound failure. (Index cleanup is expected to
+		 * finish very quickly in cases where there were no ambulkdelete()
+		 * calls.)
+		 */
+		vacrel->do_index_vacuuming = false;
+		ereport(elevel,
+				(errmsg("\"%s\": index scan bypassed: %u pages from table (%.2f%% of total) have %lld dead item identifiers",
+						vacrel->relname, vacrel->rel_pages,
+						100.0 * vacrel->lpdead_item_pages / vacrel->rel_pages,
+						(long long) vacrel->lpdead_items)));
+	}
+	else if (lazy_vacuum_all_indexes(vacrel))
 	{
 		/*
 		 * We successfully completed a round of index vacuuming.  Do related
