Don't reset pg_class.reltuples and relpages in VACUUM, if any pages were

skipped. We could update relpages anyway, but it seems better to only
update it together with reltuples, because we use the reltuples/relpages
ratio in the planner. Also don't update n_live_tuples in pgstat.

ANALYZE in VACUUM ANALYZE now needs to update pg_class, if the
VACUUM-phase didn't do so. Added some boolean-passing to let analyze_rel
know if it should update pg_class or not.

I also moved the relcache invalidation (to update rd_targblock) from
vac_update_relstats to where RelationTruncate is called, because
vac_update_relstats is not called for partial vacuums anymore. It's more
obvious to send the invalidation close to the truncation that requires it.

Per report by Ned T. Crigler.

Author: hlinnaka

src/backend/commands/analyze.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/analyze.c,v 1.129 2008/12/13 19:13:44 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/analyze.c,v 1.130 2008/12/17 09:15:02 heikki Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -101,10 +101,18 @@ static bool std_typanalyze(VacAttrStats *stats);
 
 /*
  *	analyze_rel() -- analyze one relation
+ *
+ * If update_reltuples is true, we update reltuples and relpages columns
+ * in pg_class.  Caller should pass false if we're part of VACUUM ANALYZE,
+ * and the VACUUM didn't skip any pages.  We only have an approximate count,
+ * so we don't want to overwrite the accurate values already inserted by the
+ * VACUUM in that case.  VACUUM always scans all indexes, however, so the
+ * pg_class entries for indexes are never updated if we're part of VACUUM
+ * ANALYZE.
  */
 void
 analyze_rel(Oid relid, VacuumStmt *vacstmt,
-			BufferAccessStrategy bstrategy)
+			BufferAccessStrategy bstrategy, bool update_reltuples)
 {
 	Relation	onerel;
 	int			attr_cnt,
@@ -364,7 +372,7 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt,
 		 * autovacuum code doesn't go nuts trying to get stats about a
 		 * zero-column table.
 		 */
-		if (!vacstmt->vacuum)
+		if (update_reltuples)
 			pgstat_report_analyze(onerel, 0, 0);
 		goto cleanup;
 	}
@@ -455,18 +463,24 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt,
 	}
 
 	/*
-	 * If we are running a standalone ANALYZE, update pages/tuples stats in
-	 * pg_class.  We know the accurate page count from the smgr, but only an
-	 * approximate number of tuples; therefore, if we are part of VACUUM
-	 * ANALYZE do *not* overwrite the accurate count already inserted by
-	 * VACUUM.	The same consideration applies to indexes.
+	 * Update pages/tuples stats in pg_class.
 	 */
-	if (!vacstmt->vacuum)
+	if (update_reltuples)
 	{
 		vac_update_relstats(onerel,
 							RelationGetNumberOfBlocks(onerel),
 							totalrows, hasindex, InvalidTransactionId);
+		/* report results to the stats collector, too */
+		pgstat_report_analyze(onerel, totalrows, totaldeadrows);
+	}
 
+	/*
+	 * Same for indexes. Vacuum always scans all indexes, so if we're part of
+	 * VACUUM ANALYZE, don't overwrite the accurate count already inserted by 
+	 * VACUUM.
+	 */
+	if (!vacstmt->vacuum)
+	{
 		for (ind = 0; ind < nindexes; ind++)
 		{
 			AnlIndexData *thisdata = &indexdata[ind];
@@ -477,9 +491,6 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt,
 								RelationGetNumberOfBlocks(Irel[ind]),
 								totalindexrows, false, InvalidTransactionId);
 		}
-
-		/* report results to the stats collector, too */
-		pgstat_report_analyze(onerel, totalrows, totaldeadrows);
 	}
 
 	/* We skip to here if there were no analyzable columns */

src/backend/commands/vacuum.c

@@ -13,7 +13,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/vacuum.c,v 1.382 2008/12/03 13:05:22 heikki Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/vacuum.c,v 1.383 2008/12/17 09:15:02 heikki Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -216,7 +216,7 @@ static List *get_rel_oids(Oid relid, const RangeVar *vacrel,
 			 const char *stmttype);
 static void vac_truncate_clog(TransactionId frozenXID);
 static void vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast,
-		   bool for_wraparound);
+		   bool for_wraparound, bool *scanned_all);
 static void full_vacuum_rel(Relation onerel, VacuumStmt *vacstmt);
 static void scan_heap(VRelStats *vacrelstats, Relation onerel,
 		  VacPageList vacuum_pages, VacPageList fraged_pages);
@@ -436,9 +436,11 @@ vacuum(VacuumStmt *vacstmt, Oid relid, bool do_toast,
 		foreach(cur, relations)
 		{
 			Oid			relid = lfirst_oid(cur);
+			bool		scanned_all = false;
 
 			if (vacstmt->vacuum)
-				vacuum_rel(relid, vacstmt, do_toast, for_wraparound);
+				vacuum_rel(relid, vacstmt, do_toast, for_wraparound,
+						   &scanned_all);
 
 			if (vacstmt->analyze)
 			{
@@ -460,7 +462,7 @@ vacuum(VacuumStmt *vacstmt, Oid relid, bool do_toast,
 				else
 					old_context = MemoryContextSwitchTo(anl_context);
 
-				analyze_rel(relid, vacstmt, vac_strategy);
+				analyze_rel(relid, vacstmt, vac_strategy, !scanned_all);
 
 				if (use_own_xacts)
 				{
@@ -756,21 +758,9 @@ vac_update_relstats(Relation relation,
 		dirty = true;
 	}
 
-	/*
-	 * If anything changed, write out the tuple.  Even if nothing changed,
-	 * force relcache invalidation so all backends reset their rd_targblock
-	 * --- otherwise it might point to a page we truncated away.
-	 */
+	/* If anything changed, write out the tuple. */
 	if (dirty)
-	{
 		heap_inplace_update(rd, ctup);
-		/* the above sends a cache inval message */
-	}
-	else
-	{
-		/* no need to change tuple, but force relcache inval anyway */
-		CacheInvalidateRelcacheByTuple(ctup);
-	}
 
 	heap_close(rd, RowExclusiveLock);
 }
@@ -986,10 +976,14 @@ vac_truncate_clog(TransactionId frozenXID)
  *		many small transactions.  Otherwise, two-phase locking would require
  *		us to lock the entire database during one pass of the vacuum cleaner.
  *
+ *		We'll return true in *scanned_all if the vacuum scanned all heap
+ *		pages, and updated pg_class.
+ *
  *		At entry and exit, we are not inside a transaction.
  */
 static void
-vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
+vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound,
+		   bool *scanned_all)
 {
 	LOCKMODE	lmode;
 	Relation	onerel;
@@ -998,6 +992,9 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 	Oid			save_userid;
 	bool		save_secdefcxt;
 
+	if (scanned_all)
+		*scanned_all = false;
+
 	/* Begin a transaction for vacuuming this relation */
 	StartTransactionCommand();
 
@@ -1162,7 +1159,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 	if (vacstmt->full)
 		full_vacuum_rel(onerel, vacstmt);
 	else
-		lazy_vacuum_rel(onerel, vacstmt, vac_strategy);
+		lazy_vacuum_rel(onerel, vacstmt, vac_strategy, scanned_all);
 
 	/* Restore userid */
 	SetUserIdAndContext(save_userid, save_secdefcxt);
@@ -1184,7 +1181,7 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 	 * totally unimportant for toast relations.
 	 */
 	if (toast_relid != InvalidOid)
-		vacuum_rel(toast_relid, vacstmt, false, for_wraparound);
+		vacuum_rel(toast_relid, vacstmt, false, for_wraparound, NULL);
 
 	/*
 	 * Now release the session-level lock on the master table.
@@ -1296,7 +1293,7 @@ full_vacuum_rel(Relation onerel, VacuumStmt *vacstmt)
 
 	/* report results to the stats collector, too */
 	pgstat_report_vacuum(RelationGetRelid(onerel), onerel->rd_rel->relisshared,
-						 vacstmt->analyze, vacrelstats->rel_tuples);
+						 true, vacstmt->analyze, vacrelstats->rel_tuples);
 }
 
 
@@ -2866,6 +2863,10 @@ repair_frag(VRelStats *vacrelstats, Relation onerel,
 	if (blkno < nblocks)
 	{
 		RelationTruncate(onerel, blkno);
+
+		/* force relcache inval so all backends reset their rd_targblock */
+		CacheInvalidateRelcache(onerel);
+
 		vacrelstats->rel_pages = blkno; /* set new number of blocks */
 	}
 
@@ -3286,6 +3287,10 @@ vacuum_heap(VRelStats *vacrelstats, Relation onerel, VacPageList vacuum_pages)
 						RelationGetRelationName(onerel),
 						vacrelstats->rel_pages, relblocks)));
 		RelationTruncate(onerel, relblocks);
+
+		/* force relcache inval so all backends reset their rd_targblock */
+		CacheInvalidateRelcache(onerel);
+
 		vacrelstats->rel_pages = relblocks;		/* set new number of blocks */
 	}
 }

src/backend/commands/vacuumlazy.c

@@ -29,7 +29,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/vacuumlazy.c,v 1.113 2008/12/04 11:42:23 heikki Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/vacuumlazy.c,v 1.114 2008/12/17 09:15:02 heikki Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -50,6 +50,7 @@
 #include "storage/bufmgr.h"
 #include "storage/freespace.h"
 #include "storage/lmgr.h"
+#include "utils/inval.h"
 #include "utils/lsyscache.h"
 #include "utils/memutils.h"
 #include "utils/pg_rusage.h"
@@ -135,7 +136,7 @@ static int	vac_cmp_itemptr(const void *left, const void *right);
  */
 void
 lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt,
-				BufferAccessStrategy bstrategy)
+				BufferAccessStrategy bstrategy, bool *scanned_all)
 {
 	LVRelStats *vacrelstats;
 	Relation   *Irel;
@@ -163,7 +164,7 @@ lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt,
 	vacrelstats = (LVRelStats *) palloc0(sizeof(LVRelStats));
 
 	vacrelstats->num_index_scans = 0;
-	vacrelstats->scanned_all = true;
+	vacrelstats->scanned_all = true; /* will be cleared if we skip a page */
 
 	/* Open all indexes of the relation */
 	vac_open_indexes(onerel, RowExclusiveLock, &nindexes, &Irel);
@@ -190,16 +191,23 @@ lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt,
 	FreeSpaceMapVacuum(onerel);
 
 	/*
-	 * Update statistics in pg_class. We can only advance relfrozenxid if we
-	 * didn't skip any pages.
+	 * Update statistics in pg_class.  But only if we didn't skip any pages;
+	 * the tuple count only includes tuples from the pages we've visited, and
+	 * we haven't frozen tuples in unvisited pages either.  The page count is
+	 * accurate in any case, but because we use the reltuples / relpages
+	 * ratio in the planner, it's better to not update relpages either if we
+	 * can't update reltuples.
 	 */
-	vac_update_relstats(onerel,
-						vacrelstats->rel_pages, vacrelstats->rel_tuples,
-						vacrelstats->hasindex,
-						vacrelstats->scanned_all ? FreezeLimit : InvalidOid);
+	if (vacrelstats->scanned_all)
+		vac_update_relstats(onerel,
+							vacrelstats->rel_pages, vacrelstats->rel_tuples,
+							vacrelstats->hasindex,
+							FreezeLimit);
 
 	/* report results to the stats collector, too */
-	pgstat_report_vacuum(RelationGetRelid(onerel), onerel->rd_rel->relisshared,
+	pgstat_report_vacuum(RelationGetRelid(onerel),
+						 onerel->rd_rel->relisshared,
+						 vacrelstats->scanned_all,
 						 vacstmt->analyze, vacrelstats->rel_tuples);
 
 	/* and log the action if appropriate */
@@ -221,6 +229,9 @@ lazy_vacuum_rel(Relation onerel, VacuumStmt *vacstmt,
 						vacrelstats->tuples_deleted, vacrelstats->rel_tuples,
 							pg_rusage_show(&ru0))));
 	}
+
+	if (scanned_all)
+		*scanned_all = vacrelstats->scanned_all;
 }
 
 
@@ -952,12 +963,14 @@ lazy_truncate_heap(Relation onerel, LVRelStats *vacrelstats)
 	 */
 	RelationTruncate(onerel, new_rel_pages);
 
+	/* force relcache inval so all backends reset their rd_targblock */
+	CacheInvalidateRelcache(onerel);
+
 	/*
 	 * Note: once we have truncated, we *must* keep the exclusive lock until
-	 * commit.	The sinval message that will be sent at commit (as a result of
-	 * vac_update_relstats()) must be received by other backends, to cause
-	 * them to reset their rd_targblock values, before they can safely access
-	 * the table again.
+	 * commit.	The sinval message won't be sent until commit, and other
+	 * backends must see it and reset their rd_targblock values before they
+	 * can safely access the table again.
 	 */
 
 	/* update statistics */

src/backend/parser/gram.y

@@ -11,7 +11,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/parser/gram.y,v 2.643 2008/12/04 17:51:26 petere Exp $
+ *	  $PostgreSQL: pgsql/src/backend/parser/gram.y,v 2.644 2008/12/17 09:15:02 heikki Exp $
  *
  * HISTORY
  *	  AUTHOR			DATE			MAJOR EVENT
@@ -5875,7 +5875,7 @@ VacuumStmt: VACUUM opt_full opt_freeze opt_verbose
 					n->analyze = false;
 					n->full = $2;
 					n->freeze_min_age = $3 ? 0 : -1;
-					n->scan_all = $3;
+					n->scan_all = $2 || $3;
 					n->verbose = $4;
 					n->relation = NULL;
 					n->va_cols = NIL;
@@ -5888,7 +5888,7 @@ VacuumStmt: VACUUM opt_full opt_freeze opt_verbose
 					n->analyze = false;
 					n->full = $2;
 					n->freeze_min_age = $3 ? 0 : -1;
-					n->scan_all = $3;
+					n->scan_all = $2 || $3;
 					n->verbose = $4;
 					n->relation = $5;
 					n->va_cols = NIL;
@@ -5900,7 +5900,7 @@ VacuumStmt: VACUUM opt_full opt_freeze opt_verbose
 					n->vacuum = true;
 					n->full = $2;
 					n->freeze_min_age = $3 ? 0 : -1;
-					n->scan_all = $3;
+					n->scan_all = $2 || $3;
 					n->verbose |= $4;
 					$$ = (Node *)n;
 				}

src/backend/postmaster/pgstat.c

@@ -13,7 +13,7 @@
  *
  *	Copyright (c) 2001-2008, PostgreSQL Global Development Group
  *
- *	$PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.185 2008/12/08 15:44:54 tgl Exp $
+ *	$PostgreSQL: pgsql/src/backend/postmaster/pgstat.c,v 1.186 2008/12/17 09:15:03 heikki Exp $
  * ----------
  */
 #include "postgres.h"
@@ -1178,7 +1178,7 @@ pgstat_report_autovac(Oid dboid)
  * ---------
  */
 void
-pgstat_report_vacuum(Oid tableoid, bool shared,
+pgstat_report_vacuum(Oid tableoid, bool shared, bool scanned_all,
 					 bool analyze, PgStat_Counter tuples)
 {
 	PgStat_MsgVacuum msg;
@@ -1189,6 +1189,7 @@ pgstat_report_vacuum(Oid tableoid, bool shared,
 	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_VACUUM);
 	msg.m_databaseid = shared ? InvalidOid : MyDatabaseId;
 	msg.m_tableoid = tableoid;
+	msg.m_scanned_all = scanned_all;
 	msg.m_analyze = analyze;
 	msg.m_autovacuum = IsAutoVacuumWorkerProcess();		/* is this autovacuum? */
 	msg.m_vacuumtime = GetCurrentTimestamp();
@@ -3765,12 +3766,14 @@ pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
 		tabentry->autovac_vacuum_timestamp = msg->m_vacuumtime;
 	else
 		tabentry->vacuum_timestamp = msg->m_vacuumtime;
-	tabentry->n_live_tuples = msg->m_tuples;
+	if (msg->m_scanned_all)
+		tabentry->n_live_tuples = msg->m_tuples;
 	/* Resetting dead_tuples to 0 is an approximation ... */
 	tabentry->n_dead_tuples = 0;
 	if (msg->m_analyze)
 	{
-		tabentry->last_anl_tuples = msg->m_tuples;
+		if (msg->m_scanned_all)
+			tabentry->last_anl_tuples = msg->m_tuples;
 		if (msg->m_autovacuum)
 			tabentry->autovac_analyze_timestamp = msg->m_vacuumtime;
 		else
@@ -3780,7 +3783,7 @@ pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len)
 	{
 		/* last_anl_tuples must never exceed n_live_tuples+n_dead_tuples */
 		tabentry->last_anl_tuples = Min(tabentry->last_anl_tuples,
-										msg->m_tuples);
+										tabentry->n_live_tuples);
 	}
 }