Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit37fa3b6

Browse files
committed
Tweak indexscan and seqscan code to arrange that steps from one page to
the next are handled by ReleaseAndReadBuffer rather than separateReleaseBuffer and ReadBuffer calls. This cuts the number of acquisitionsof the BufMgrLock by a factor of 2 (possibly more, if an indexscan happensto pull successive rows from the same heap page). Unfortunately thisdoesn't seem enough to get us out of the recently discussed context-switchstorm problem, but it's surely worth doing anyway.
1 parent95a03e9 commit37fa3b6

File tree

11 files changed

+132
-65
lines changed

11 files changed

+132
-65
lines changed

‎src/backend/access/heap/heapam.c

Lines changed: 26 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.164 2004/04/01 21:28:43 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/heap/heapam.c,v 1.165 2004/04/21 18:24:23 tgl Exp $
1212
*
1313
*
1414
* INTERFACE ROUTINES
@@ -888,6 +888,28 @@ heap_fetch(Relation relation,
888888
Buffer*userbuf,
889889
boolkeep_buf,
890890
PgStat_Info*pgstat_info)
891+
{
892+
/* Assume *userbuf is undefined on entry */
893+
*userbuf=InvalidBuffer;
894+
returnheap_release_fetch(relation,snapshot,tuple,
895+
userbuf,keep_buf,pgstat_info);
896+
}
897+
898+
/*
899+
*heap_release_fetch- retrieve tuple with given tid
900+
*
901+
* This has the same API as heap_fetch except that if *userbuf is not
902+
* InvalidBuffer on entry, that buffer will be released before reading
903+
* the new page. This saves a separate ReleaseBuffer step and hence
904+
* one entry into the bufmgr when looping through multiple fetches.
905+
*/
906+
bool
907+
heap_release_fetch(Relationrelation,
908+
Snapshotsnapshot,
909+
HeapTupletuple,
910+
Buffer*userbuf,
911+
boolkeep_buf,
912+
PgStat_Info*pgstat_info)
891913
{
892914
ItemPointertid=&(tuple->t_self);
893915
ItemIdlp;
@@ -898,9 +920,10 @@ heap_fetch(Relation relation,
898920

899921
/*
900922
* get the buffer from the relation descriptor. Note that this does a
901-
* buffer pin.
923+
* buffer pin, and releases the old *userbuf if not InvalidBuffer.
902924
*/
903-
buffer=ReadBuffer(relation,ItemPointerGetBlockNumber(tid));
925+
buffer=ReleaseAndReadBuffer(*userbuf,relation,
926+
ItemPointerGetBlockNumber(tid));
904927

905928
if (!BufferIsValid(buffer))
906929
elog(ERROR,"ReadBuffer(\"%s\", %lu) failed",

‎src/backend/access/index/indexam.c

Lines changed: 20 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/index/indexam.c,v 1.72 2003/11/29 19:51:40 pgsql Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/index/indexam.c,v 1.73 2004/04/21 18:24:24 tgl Exp $
1212
*
1313
* INTERFACE ROUTINES
1414
*index_open- open an index relation by relation OID
@@ -316,6 +316,13 @@ index_rescan(IndexScanDesc scan, ScanKey key)
316316
SCAN_CHECKS;
317317
GET_SCAN_PROCEDURE(rescan,amrescan);
318318

319+
/* Release any held pin on a heap page */
320+
if (BufferIsValid(scan->xs_cbuf))
321+
{
322+
ReleaseBuffer(scan->xs_cbuf);
323+
scan->xs_cbuf=InvalidBuffer;
324+
}
325+
319326
scan->kill_prior_tuple= false;/* for safety */
320327
scan->keys_are_unique= false;/* may be set by index AM */
321328
scan->got_tuple= false;
@@ -419,13 +426,6 @@ index_getnext(IndexScanDesc scan, ScanDirection direction)
419426

420427
SCAN_CHECKS;
421428

422-
/* Release any previously held pin */
423-
if (BufferIsValid(scan->xs_cbuf))
424-
{
425-
ReleaseBuffer(scan->xs_cbuf);
426-
scan->xs_cbuf=InvalidBuffer;
427-
}
428-
429429
/*
430430
* If we already got a tuple and it must be unique, there's no need to
431431
* make the index AM look through any additional tuples. (This can
@@ -508,14 +508,22 @@ index_getnext(IndexScanDesc scan, ScanDirection direction)
508508
scan->kill_prior_tuple= false;
509509

510510
if (!found)
511+
{
512+
/* Release any held pin on a heap page */
513+
if (BufferIsValid(scan->xs_cbuf))
514+
{
515+
ReleaseBuffer(scan->xs_cbuf);
516+
scan->xs_cbuf=InvalidBuffer;
517+
}
511518
returnNULL;/* failure exit */
519+
}
512520

513521
/*
514522
* Fetch the heap tuple and see if it matches the snapshot.
515523
*/
516-
if (heap_fetch(scan->heapRelation,scan->xs_snapshot,
517-
heapTuple,&scan->xs_cbuf, true,
518-
&scan->xs_pgstat_info))
524+
if (heap_release_fetch(scan->heapRelation,scan->xs_snapshot,
525+
heapTuple,&scan->xs_cbuf, true,
526+
&scan->xs_pgstat_info))
519527
break;
520528

521529
/* Skip if no tuple at this location */
@@ -527,7 +535,7 @@ index_getnext(IndexScanDesc scan, ScanDirection direction)
527535
* if the tuple is dead to all transactions. If so, signal the
528536
* index AM to not return it on future indexscans.
529537
*
530-
* We toldheap_fetch to keep a pin on the buffer, so we can
538+
* We toldheap_release_fetch to keep a pin on the buffer, so we can
531539
* re-access the tuple here. But we must re-lock the buffer
532540
* first. Also, it's just barely possible for an update of hint
533541
* bits to occur here.
@@ -542,8 +550,6 @@ index_getnext(IndexScanDesc scan, ScanDirection direction)
542550
if (sv_infomask!=heapTuple->t_data->t_infomask)
543551
SetBufferCommitInfoNeedsSave(scan->xs_cbuf);
544552
LockBuffer(scan->xs_cbuf,BUFFER_LOCK_UNLOCK);
545-
ReleaseBuffer(scan->xs_cbuf);
546-
scan->xs_cbuf=InvalidBuffer;
547553
}
548554

549555
/* Success exit */

‎src/backend/access/nbtree/nbtinsert.c

Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
*
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.111 2004/01/07 18:56:24 neilc Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtinsert.c,v 1.112 2004/04/21 18:24:25 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -297,9 +297,7 @@ _bt_check_unique(Relation rel, BTItem btitem, Relation heapRel,
297297
for (;;)
298298
{
299299
nblkno=opaque->btpo_next;
300-
if (nbuf!=InvalidBuffer)
301-
_bt_relbuf(rel,nbuf);
302-
nbuf=_bt_getbuf(rel,nblkno,BT_READ);
300+
nbuf=_bt_relandgetbuf(rel,nbuf,nblkno,BT_READ);
303301
page=BufferGetPage(nbuf);
304302
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
305303
if (!P_IGNORE(opaque))
@@ -454,9 +452,7 @@ _bt_insertonpg(Relation rel,
454452
{
455453
BlockNumberrblkno=lpageop->btpo_next;
456454

457-
if (rbuf!=InvalidBuffer)
458-
_bt_relbuf(rel,rbuf);
459-
rbuf=_bt_getbuf(rel,rblkno,BT_WRITE);
455+
rbuf=_bt_relandgetbuf(rel,rbuf,rblkno,BT_WRITE);
460456
page=BufferGetPage(rbuf);
461457
lpageop= (BTPageOpaque)PageGetSpecialPointer(page);
462458
if (!P_IGNORE(lpageop))
@@ -1357,7 +1353,7 @@ _bt_getstackbuf(Relation rel, BTStack stack, int access)
13571353
if (P_RIGHTMOST(opaque))
13581354
{
13591355
_bt_relbuf(rel,buf);
1360-
return(InvalidBuffer);
1356+
returnInvalidBuffer;
13611357
}
13621358
blkno=opaque->btpo_next;
13631359
start=InvalidOffsetNumber;

‎src/backend/access/nbtree/nbtpage.c

Lines changed: 35 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@
99
*
1010
*
1111
* IDENTIFICATION
12-
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtpage.c,v 1.74 2003/12/2101:23:06 tgl Exp $
12+
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtpage.c,v 1.75 2004/04/2118:24:25 tgl Exp $
1313
*
1414
*NOTES
1515
* Postgres btree pages look like ordinary relation pages.The opaque
@@ -262,11 +262,15 @@ _bt_getroot(Relation rel, int access)
262262
Assert(rootblkno!=P_NONE);
263263
rootlevel=metad->btm_fastlevel;
264264

265-
_bt_relbuf(rel,metabuf);/* done with the meta page */
265+
/*
266+
* We are done with the metapage; arrange to release it via
267+
* first _bt_relandgetbuf call
268+
*/
269+
rootbuf=metabuf;
266270

267271
for (;;)
268272
{
269-
rootbuf=_bt_getbuf(rel,rootblkno,BT_READ);
273+
rootbuf=_bt_relandgetbuf(rel,rootbuf,rootblkno,BT_READ);
270274
rootpage=BufferGetPage(rootbuf);
271275
rootopaque= (BTPageOpaque)PageGetSpecialPointer(rootpage);
272276

@@ -278,8 +282,6 @@ _bt_getroot(Relation rel, int access)
278282
elog(ERROR,"no live root page found in \"%s\"",
279283
RelationGetRelationName(rel));
280284
rootblkno=rootopaque->btpo_next;
281-
282-
_bt_relbuf(rel,rootbuf);
283285
}
284286

285287
/* Note: can't check btpo.level on deleted pages */
@@ -352,11 +354,15 @@ _bt_gettrueroot(Relation rel)
352354
rootblkno=metad->btm_root;
353355
rootlevel=metad->btm_level;
354356

355-
_bt_relbuf(rel,metabuf);/* done with the meta page */
357+
/*
358+
* We are done with the metapage; arrange to release it via
359+
* first _bt_relandgetbuf call
360+
*/
361+
rootbuf=metabuf;
356362

357363
for (;;)
358364
{
359-
rootbuf=_bt_getbuf(rel,rootblkno,BT_READ);
365+
rootbuf=_bt_relandgetbuf(rel,rootbuf,rootblkno,BT_READ);
360366
rootpage=BufferGetPage(rootbuf);
361367
rootopaque= (BTPageOpaque)PageGetSpecialPointer(rootpage);
362368

@@ -368,8 +374,6 @@ _bt_gettrueroot(Relation rel)
368374
elog(ERROR,"no live root page found in \"%s\"",
369375
RelationGetRelationName(rel));
370376
rootblkno=rootopaque->btpo_next;
371-
372-
_bt_relbuf(rel,rootbuf);
373377
}
374378

375379
/* Note: can't check btpo.level on deleted pages */
@@ -492,6 +496,28 @@ _bt_getbuf(Relation rel, BlockNumber blkno, int access)
492496
returnbuf;
493497
}
494498

499+
/*
500+
*_bt_relandgetbuf() -- release a locked buffer and get another one.
501+
*
502+
* This is equivalent to _bt_relbuf followed by _bt_getbuf, with the
503+
* exception that blkno may not be P_NEW. Also, if obuf is InvalidBuffer
504+
* then it reduces to just _bt_getbuf; allowing this case simplifies some
505+
* callers. The motivation for using this is to avoid two entries to the
506+
* bufmgr when one will do.
507+
*/
508+
Buffer
509+
_bt_relandgetbuf(Relationrel,Bufferobuf,BlockNumberblkno,intaccess)
510+
{
511+
Bufferbuf;
512+
513+
Assert(blkno!=P_NEW);
514+
if (BufferIsValid(obuf))
515+
LockBuffer(obuf,BUFFER_LOCK_UNLOCK);
516+
buf=ReleaseAndReadBuffer(obuf,rel,blkno);
517+
LockBuffer(buf,access);
518+
returnbuf;
519+
}
520+
495521
/*
496522
*_bt_relbuf() -- release a locked buffer.
497523
*

‎src/backend/access/nbtree/nbtree.c

Lines changed: 2 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,7 @@
1212
* Portions Copyright (c) 1994, Regents of the University of California
1313
*
1414
* IDENTIFICATION
15-
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.113 2004/02/10 03:42:43 tgl Exp $
15+
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.114 2004/04/21 18:24:26 tgl Exp $
1616
*
1717
*-------------------------------------------------------------------------
1818
*/
@@ -919,9 +919,7 @@ _bt_restscan(IndexScanDesc scan)
919919
for (;;)
920920
{
921921
blkno=opaque->btpo_next;
922-
if (nextbuf!=InvalidBuffer)
923-
_bt_relbuf(rel,nextbuf);
924-
nextbuf=_bt_getbuf(rel,blkno,BT_READ);
922+
nextbuf=_bt_relandgetbuf(rel,nextbuf,blkno,BT_READ);
925923
page=BufferGetPage(nextbuf);
926924
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
927925
if (!P_IGNORE(opaque))

‎src/backend/access/nbtree/nbtsearch.c

Lines changed: 10 additions & 19 deletions
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,7 @@
88
* Portions Copyright (c) 1994, Regents of the University of California
99
*
1010
* IDENTIFICATION
11-
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.86 2003/12/2117:52:34 tgl Exp $
11+
* $PostgreSQL: pgsql/src/backend/access/nbtree/nbtsearch.c,v 1.87 2004/04/2118:24:26 tgl Exp $
1212
*
1313
*-------------------------------------------------------------------------
1414
*/
@@ -108,8 +108,7 @@ _bt_search(Relation rel, int keysz, ScanKey scankey, bool nextkey,
108108
new_stack->bts_parent=stack_in;
109109

110110
/* drop the read lock on the parent page, acquire one on the child */
111-
_bt_relbuf(rel,*bufP);
112-
*bufP=_bt_getbuf(rel,blkno,BT_READ);
111+
*bufP=_bt_relandgetbuf(rel,*bufP,blkno,BT_READ);
113112

114113
/* okay, all set to move down a level */
115114
stack_in=new_stack;
@@ -178,8 +177,7 @@ _bt_moveright(Relation rel,
178177
/* step right one page */
179178
BlockNumberrblkno=opaque->btpo_next;
180179

181-
_bt_relbuf(rel,buf);
182-
buf=_bt_getbuf(rel,rblkno,access);
180+
buf=_bt_relandgetbuf(rel,buf,rblkno,access);
183181
page=BufferGetPage(buf);
184182
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
185183
}
@@ -933,8 +931,7 @@ _bt_step(IndexScanDesc scan, Buffer *bufP, ScanDirection dir)
933931
}
934932
/* step right one page */
935933
blkno=opaque->btpo_next;
936-
_bt_relbuf(rel,*bufP);
937-
*bufP=_bt_getbuf(rel,blkno,BT_READ);
934+
*bufP=_bt_relandgetbuf(rel,*bufP,blkno,BT_READ);
938935
page=BufferGetPage(*bufP);
939936
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
940937
if (!P_IGNORE(opaque))
@@ -1041,8 +1038,7 @@ _bt_walk_left(Relation rel, Buffer buf)
10411038
obknum=BufferGetBlockNumber(buf);
10421039
/* step left */
10431040
blkno=lblkno=opaque->btpo_prev;
1044-
_bt_relbuf(rel,buf);
1045-
buf=_bt_getbuf(rel,blkno,BT_READ);
1041+
buf=_bt_relandgetbuf(rel,buf,blkno,BT_READ);
10461042
page=BufferGetPage(buf);
10471043
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
10481044

@@ -1069,15 +1065,13 @@ _bt_walk_left(Relation rel, Buffer buf)
10691065
if (P_RIGHTMOST(opaque)||++tries>4)
10701066
break;
10711067
blkno=opaque->btpo_next;
1072-
_bt_relbuf(rel,buf);
1073-
buf=_bt_getbuf(rel,blkno,BT_READ);
1068+
buf=_bt_relandgetbuf(rel,buf,blkno,BT_READ);
10741069
page=BufferGetPage(buf);
10751070
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
10761071
}
10771072

10781073
/* Return to the original page to see what's up */
1079-
_bt_relbuf(rel,buf);
1080-
buf=_bt_getbuf(rel,obknum,BT_READ);
1074+
buf=_bt_relandgetbuf(rel,buf,obknum,BT_READ);
10811075
page=BufferGetPage(buf);
10821076
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
10831077
if (P_ISDELETED(opaque))
@@ -1094,8 +1088,7 @@ _bt_walk_left(Relation rel, Buffer buf)
10941088
elog(ERROR,"fell off the end of \"%s\"",
10951089
RelationGetRelationName(rel));
10961090
blkno=opaque->btpo_next;
1097-
_bt_relbuf(rel,buf);
1098-
buf=_bt_getbuf(rel,blkno,BT_READ);
1091+
buf=_bt_relandgetbuf(rel,buf,blkno,BT_READ);
10991092
page=BufferGetPage(buf);
11001093
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
11011094
if (!P_ISDELETED(opaque))
@@ -1177,8 +1170,7 @@ _bt_get_endpoint(Relation rel, uint32 level, bool rightmost)
11771170
if (blkno==P_NONE)
11781171
elog(ERROR,"fell off the end of \"%s\"",
11791172
RelationGetRelationName(rel));
1180-
_bt_relbuf(rel,buf);
1181-
buf=_bt_getbuf(rel,blkno,BT_READ);
1173+
buf=_bt_relandgetbuf(rel,buf,blkno,BT_READ);
11821174
page=BufferGetPage(buf);
11831175
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
11841176
}
@@ -1199,8 +1191,7 @@ _bt_get_endpoint(Relation rel, uint32 level, bool rightmost)
11991191
itup=&(btitem->bti_itup);
12001192
blkno=ItemPointerGetBlockNumber(&(itup->t_tid));
12011193

1202-
_bt_relbuf(rel,buf);
1203-
buf=_bt_getbuf(rel,blkno,BT_READ);
1194+
buf=_bt_relandgetbuf(rel,buf,blkno,BT_READ);
12041195
page=BufferGetPage(buf);
12051196
opaque= (BTPageOpaque)PageGetSpecialPointer(page);
12061197
}

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp