Make large sequential scans and VACUUMs work in a limited-size "ring" of

buffers, rather than blowing out the whole shared-buffer arena.  Aside from
avoiding cache spoliation, this fixes the problem that VACUUM formerly tended
to cause a WAL flush for every page it modified, because we had it hacked to
use only a single buffer.  Those flushes will now occur only once per
ring-ful.  The exact ring size, and the threshold for seqscans to switch into
the ring usage pattern, remain under debate; but the infrastructure seems
done.  The key bit of infrastructure is a new optional BufferAccessStrategy
object that can be passed to ReadBuffer operations; this replaces the former
StrategyHintVacuum API.

This patch also changes the buffer usage-count methodology a bit: we now
advance usage_count when first pinning a buffer, rather than when last
unpinning it.  To preserve the behavior that a buffer's lifetime starts to
decrease when it's released, the clock sweep code is modified to not decrement
usage_count of pinned buffers.

Work not done in this commit: teach GiST and GIN indexes to use the vacuum
BufferAccessStrategy for vacuum-driven fetches.

Original patch by Simon, reworked by Heikki and again by Tom.

src/backend/access/hash/hash.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/hash/hash.c,v 1.94 2007/05/03 16:45:58 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/hash/hash.c,v 1.95 2007/05/30 20:11:51 tgl Exp $
  *
  * NOTES
  *	  This file contains only the public interface routines.
@@ -547,8 +547,9 @@ loop_top:
 
 			vacuum_delay_point();
 
-			buf = _hash_getbuf(rel, blkno, HASH_WRITE,
-							   LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
+			buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
+											 LH_BUCKET_PAGE | LH_OVERFLOW_PAGE,
+											 info->strategy);
 			page = BufferGetPage(buf);
 			opaque = (HashPageOpaque) PageGetSpecialPointer(page);
 			Assert(opaque->hasho_bucket == cur_bucket);
@@ -596,7 +597,8 @@ loop_top:
 
 		/* If we deleted anything, try to compact free space */
 		if (bucket_dirty)
-			_hash_squeezebucket(rel, cur_bucket, bucket_blkno);
+			_hash_squeezebucket(rel, cur_bucket, bucket_blkno,
+								info->strategy);
 
 		/* Release bucket lock */
 		_hash_droplock(rel, bucket_blkno, HASH_EXCLUSIVE);

src/backend/access/hash/hashovfl.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/hash/hashovfl.c,v 1.57 2007/05/03 16:45:58 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/hash/hashovfl.c,v 1.58 2007/05/30 20:11:51 tgl Exp $
  *
  * NOTES
  *	  Overflow pages look like ordinary relation pages.
@@ -362,6 +362,9 @@ _hash_firstfreebit(uint32 map)
  *	Remove this overflow page from its bucket's chain, and mark the page as
  *	free.  On entry, ovflbuf is write-locked; it is released before exiting.
  *
+ *	Since this function is invoked in VACUUM, we provide an access strategy
+ *	parameter that controls fetches of the bucket pages.
+ *
  *	Returns the block number of the page that followed the given page
  *	in the bucket, or InvalidBlockNumber if no following page.
  *
@@ -370,7 +373,8 @@ _hash_firstfreebit(uint32 map)
  *	on the bucket, too.
  */
 BlockNumber
-_hash_freeovflpage(Relation rel, Buffer ovflbuf)
+_hash_freeovflpage(Relation rel, Buffer ovflbuf,
+				   BufferAccessStrategy bstrategy)
 {
 	HashMetaPage metap;
 	Buffer		metabuf;
@@ -413,8 +417,11 @@ _hash_freeovflpage(Relation rel, Buffer ovflbuf)
 	 */
 	if (BlockNumberIsValid(prevblkno))
 	{
-		Buffer		prevbuf = _hash_getbuf(rel, prevblkno, HASH_WRITE,
-										   LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
+		Buffer		prevbuf = _hash_getbuf_with_strategy(rel,
+														 prevblkno,
+														 HASH_WRITE,
+														 LH_BUCKET_PAGE | LH_OVERFLOW_PAGE,
+														 bstrategy);
 		Page		prevpage = BufferGetPage(prevbuf);
 		HashPageOpaque prevopaque = (HashPageOpaque) PageGetSpecialPointer(prevpage);
 
@@ -424,8 +431,11 @@ _hash_freeovflpage(Relation rel, Buffer ovflbuf)
 	}
 	if (BlockNumberIsValid(nextblkno))
 	{
-		Buffer		nextbuf = _hash_getbuf(rel, nextblkno, HASH_WRITE,
-										   LH_OVERFLOW_PAGE);
+		Buffer		nextbuf = _hash_getbuf_with_strategy(rel,
+														 nextblkno,
+														 HASH_WRITE,
+														 LH_OVERFLOW_PAGE,
+														 bstrategy);
 		Page		nextpage = BufferGetPage(nextbuf);
 		HashPageOpaque nextopaque = (HashPageOpaque) PageGetSpecialPointer(nextpage);
 
@@ -434,6 +444,8 @@ _hash_freeovflpage(Relation rel, Buffer ovflbuf)
 		_hash_wrtbuf(rel, nextbuf);
 	}
 
+	/* Note: bstrategy is intentionally not used for metapage and bitmap */
+
 	/* Read the metapage so we can determine which bitmap page to use */
 	metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
 	metap = (HashMetaPage) BufferGetPage(metabuf);
@@ -558,11 +570,15 @@ _hash_initbitmap(Relation rel, HashMetaPage metap, BlockNumber blkno)
  *
  *	Caller must hold exclusive lock on the target bucket.  This allows
  *	us to safely lock multiple pages in the bucket.
+ *
+ *	Since this function is invoked in VACUUM, we provide an access strategy
+ *	parameter that controls fetches of the bucket pages.
  */
 void
 _hash_squeezebucket(Relation rel,
 					Bucket bucket,
-					BlockNumber bucket_blkno)
+					BlockNumber bucket_blkno,
+					BufferAccessStrategy bstrategy)
 {
 	Buffer		wbuf;
 	Buffer		rbuf = 0;
@@ -581,7 +597,11 @@ _hash_squeezebucket(Relation rel,
 	 * start squeezing into the base bucket page.
 	 */
 	wblkno = bucket_blkno;
-	wbuf = _hash_getbuf(rel, wblkno, HASH_WRITE, LH_BUCKET_PAGE);
+	wbuf = _hash_getbuf_with_strategy(rel,
+									  wblkno,
+									  HASH_WRITE,
+									  LH_BUCKET_PAGE,
+									  bstrategy);
 	wpage = BufferGetPage(wbuf);
 	wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
 
@@ -595,8 +615,10 @@ _hash_squeezebucket(Relation rel,
 	}
 
 	/*
-	 * find the last page in the bucket chain by starting at the base bucket
-	 * page and working forward.
+	 * Find the last page in the bucket chain by starting at the base bucket
+	 * page and working forward.  Note: we assume that a hash bucket chain is
+	 * usually smaller than the buffer ring being used by VACUUM, else using
+	 * the access strategy here would be counterproductive.
 	 */
 	ropaque = wopaque;
 	do
@@ -604,7 +626,11 @@ _hash_squeezebucket(Relation rel,
 		rblkno = ropaque->hasho_nextblkno;
 		if (ropaque != wopaque)
 			_hash_relbuf(rel, rbuf);
-		rbuf = _hash_getbuf(rel, rblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
+		rbuf = _hash_getbuf_with_strategy(rel,
+										  rblkno,
+										  HASH_WRITE,
+										  LH_OVERFLOW_PAGE,
+										  bstrategy);
 		rpage = BufferGetPage(rbuf);
 		ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
 		Assert(ropaque->hasho_bucket == bucket);
@@ -644,7 +670,11 @@ _hash_squeezebucket(Relation rel,
 					return;
 				}
 
-				wbuf = _hash_getbuf(rel, wblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
+				wbuf = _hash_getbuf_with_strategy(rel,
+												  wblkno,
+												  HASH_WRITE,
+												  LH_OVERFLOW_PAGE,
+												  bstrategy);
 				wpage = BufferGetPage(wbuf);
 				wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
 				Assert(wopaque->hasho_bucket == bucket);
@@ -688,15 +718,19 @@ _hash_squeezebucket(Relation rel,
 				/* yes, so release wbuf lock first */
 				_hash_wrtbuf(rel, wbuf);
 				/* free this overflow page (releases rbuf) */
-				_hash_freeovflpage(rel, rbuf);
+				_hash_freeovflpage(rel, rbuf, bstrategy);
 				/* done */
 				return;
 			}
 
 			/* free this overflow page, then get the previous one */
-			_hash_freeovflpage(rel, rbuf);
+			_hash_freeovflpage(rel, rbuf, bstrategy);
 
-			rbuf = _hash_getbuf(rel, rblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
+			rbuf = _hash_getbuf_with_strategy(rel,
+											  rblkno,
+											  HASH_WRITE,
+											  LH_OVERFLOW_PAGE,
+											  bstrategy);
 			rpage = BufferGetPage(rbuf);
 			ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
 			Assert(ropaque->hasho_bucket == bucket);

src/backend/access/hash/hashpage.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/hash/hashpage.c,v 1.67 2007/05/03 16:45:58 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/hash/hashpage.c,v 1.68 2007/05/30 20:11:51 tgl Exp $
  *
  * NOTES
  *	  Postgres hash pages look like ordinary relation pages.  The opaque
@@ -214,6 +214,34 @@ _hash_getnewbuf(Relation rel, BlockNumber blkno)
 	return buf;
 }
 
+/*
+ *	_hash_getbuf_with_strategy() -- Get a buffer with nondefault strategy.
+ *
+ *		This is identical to _hash_getbuf() but also allows a buffer access
+ *		strategy to be specified.  We use this for VACUUM operations.
+ */
+Buffer
+_hash_getbuf_with_strategy(Relation rel, BlockNumber blkno,
+						   int access, int flags,
+						   BufferAccessStrategy bstrategy)
+{
+	Buffer		buf;
+
+	if (blkno == P_NEW)
+		elog(ERROR, "hash AM does not use P_NEW");
+
+	buf = ReadBufferWithStrategy(rel, blkno, bstrategy);
+
+	if (access != HASH_NOLOCK)
+		LockBuffer(buf, access);
+
+	/* ref count and lock type are correct */
+
+	_hash_checkpage(rel, buf, flags);
+
+	return buf;
+}
+
 /*
  *	_hash_relbuf() -- release a locked buffer.
  *
@@ -840,5 +868,5 @@ _hash_splitbucket(Relation rel,
 	_hash_wrtbuf(rel, obuf);
 	_hash_wrtbuf(rel, nbuf);
 
-	_hash_squeezebucket(rel, obucket, start_oblkno);
+	_hash_squeezebucket(rel, obucket, start_oblkno, NULL);
 }

src/backend/access/heap/heapam.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.233 2007/05/27 03:50:38 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.234 2007/05/30 20:11:53 tgl Exp $
  *
  *
  * INTERFACE ROUTINES
@@ -83,6 +83,24 @@ initscan(HeapScanDesc scan, ScanKey key)
 	 */
 	scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_rd);
 
+	/*
+	 * If the table is large relative to NBuffers, use a bulk-read access
+	 * strategy, else use the default random-access strategy.  During a
+	 * rescan, don't make a new strategy object if we don't have to.
+	 */
+	if (scan->rs_nblocks > NBuffers / 4 &&
+		!scan->rs_rd->rd_istemp)
+	{
+		if (scan->rs_strategy == NULL)
+			scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
+	}
+	else
+	{
+		if (scan->rs_strategy != NULL)
+			FreeAccessStrategy(scan->rs_strategy);
+		scan->rs_strategy = NULL;
+	}
+
 	scan->rs_inited = false;
 	scan->rs_ctup.t_data = NULL;
 	ItemPointerSetInvalid(&scan->rs_ctup.t_self);
@@ -123,9 +141,17 @@ heapgetpage(HeapScanDesc scan, BlockNumber page)
 
 	Assert(page < scan->rs_nblocks);
 
-	scan->rs_cbuf = ReleaseAndReadBuffer(scan->rs_cbuf,
-										 scan->rs_rd,
-										 page);
+	/* release previous scan buffer, if any */
+	if (BufferIsValid(scan->rs_cbuf))
+	{
+		ReleaseBuffer(scan->rs_cbuf);
+		scan->rs_cbuf = InvalidBuffer;
+	}
+
+	/* read page using selected strategy */
+	scan->rs_cbuf = ReadBufferWithStrategy(scan->rs_rd,
+										   page,
+										   scan->rs_strategy);
 	scan->rs_cblock = page;
 
 	if (!scan->rs_pageatatime)
@@ -938,6 +964,7 @@ heap_beginscan(Relation relation, Snapshot snapshot,
 	scan->rs_rd = relation;
 	scan->rs_snapshot = snapshot;
 	scan->rs_nkeys = nkeys;
+	scan->rs_strategy = NULL;	/* set in initscan */
 
 	/*
 	 * we can use page-at-a-time mode if it's an MVCC-safe snapshot
@@ -1007,6 +1034,9 @@ heap_endscan(HeapScanDesc scan)
 	if (scan->rs_key)
 		pfree(scan->rs_key);
 
+	if (scan->rs_strategy != NULL)
+		FreeAccessStrategy(scan->rs_strategy);
+
 	pfree(scan);
 }
 

src/backend/access/nbtree/nbtree.c

@@ -12,7 +12,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.154 2007/01/05 22:19:23 momjian Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.155 2007/05/30 20:11:53 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -786,9 +786,10 @@ restart:
 	/*
 	 * We can't use _bt_getbuf() here because it always applies
 	 * _bt_checkpage(), which will barf on an all-zero page. We want to
-	 * recycle all-zero pages, not fail.
+	 * recycle all-zero pages, not fail.  Also, we want to use a nondefault
+	 * buffer access strategy.
 	 */
-	buf = ReadBuffer(rel, blkno);
+	buf = ReadBufferWithStrategy(rel, blkno, info->strategy);
 	LockBuffer(buf, BT_READ);
 	page = BufferGetPage(buf);
 	opaque = (BTPageOpaque) PageGetSpecialPointer(page);

src/backend/access/transam/xlog.c

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2007, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.269 2007/05/20 21:08:19 tgl Exp $
+ * $PostgreSQL: pgsql/src/backend/access/transam/xlog.c,v 1.270 2007/05/30 20:11:55 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1799,6 +1799,36 @@ XLogFlush(XLogRecPtr record)
 			 LogwrtResult.Flush.xlogid, LogwrtResult.Flush.xrecoff);
 }
 
+/*
+ * Test whether XLOG data has been flushed up to (at least) the given position.
+ *
+ * Returns true if a flush is still needed.  (It may be that someone else
+ * is already in process of flushing that far, however.)
+ */
+bool
+XLogNeedsFlush(XLogRecPtr record)
+{
+	/* Quick exit if already known flushed */
+	if (XLByteLE(record, LogwrtResult.Flush))
+		return false;
+
+	/* read LogwrtResult and update local state */
+	{
+		/* use volatile pointer to prevent code rearrangement */
+		volatile XLogCtlData *xlogctl = XLogCtl;
+
+		SpinLockAcquire(&xlogctl->info_lck);
+		LogwrtResult = xlogctl->LogwrtResult;
+		SpinLockRelease(&xlogctl->info_lck);
+	}
+
+	/* check again */
+	if (XLByteLE(record, LogwrtResult.Flush))
+		return false;
+
+	return true;
+}
+
 /*
  * Create a new XLOG file segment, or open a pre-existing one.
  *