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

Commit1500e26

Browse files
author
Hiroshi Inoue
committed
Fix for TODO item * spinlock stuck problem when elog(FATAL)and elog(ERROR) inside bufmgr.
1 parent116ba5d commit1500e26

File tree

3 files changed

+171
-48
lines changed

3 files changed

+171
-48
lines changed

‎src/backend/storage/buffer/bufmgr.c

Lines changed: 166 additions & 45 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
*
88
*
99
* IDENTIFICATION
10-
* $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.70 2000/01/15 02:59:33 petere Exp $
10+
* $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.71 2000/01/17 01:15:17 inoue Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -74,6 +74,18 @@ static intWriteMode = BUFFER_LATE_WRITE;/* Delayed write is
7474

7575
staticvoidWaitIO(BufferDesc*buf,SPINLOCKspinlock);
7676

77+
staticvoidStartBufferIO(BufferDesc*buf,boolforInput);
78+
staticvoidTerminateBufferIO(BufferDesc*buf);
79+
staticvoidContinueBufferIO(BufferDesc*buf,boolforInput);
80+
externvoidInitBufferIO(void);
81+
externvoidAbortBufferIO(void);
82+
83+
/*
84+
* Macro : BUFFER_IS_BROKEN
85+
* Note that write error doesn't mean the buffer broken
86+
*/
87+
#defineBUFFER_IS_BROKEN(buf) ((buf->flags & BM_IO_ERROR) && !(buf->flags & BM_DIRTY))
88+
7789
#ifndefHAS_TEST_AND_SET
7890
staticvoidSignalIO(BufferDesc*buf);
7991
externlong*NWaitIOBackendP;/* defined in buf_init.c */
@@ -312,12 +324,7 @@ ReadBufferWithBufferLock(Relation reln,
312324
}
313325

314326
/* If anyone was waiting for IO to complete, wake them up now */
315-
#ifdefHAS_TEST_AND_SET
316-
S_UNLOCK(&(bufHdr->io_in_progress_lock));
317-
#else
318-
if (bufHdr->refcount>1)
319-
SignalIO(bufHdr);
320-
#endif
327+
TerminateBufferIO(bufHdr);
321328

322329
SpinRelease(BufMgrLock);
323330

@@ -375,13 +382,19 @@ BufferAlloc(Relation reln,
375382
inProgress= (buf->flags&BM_IO_IN_PROGRESS);
376383

377384
*foundPtr= TRUE;
378-
if (inProgress)
385+
if (inProgress)/* confirm end of IO */
379386
{
380387
WaitIO(buf,BufMgrLock);
381-
if (buf->flags&BM_IO_ERROR)
382-
{
388+
inProgress= (buf->flags&BM_IO_IN_PROGRESS);
389+
}
390+
if (BUFFER_IS_BROKEN(buf))
391+
{
392+
/* I couldn't understand the following old comment.
393+
* If there's no IO for the buffer and the buffer
394+
* is BROKEN,it should be read again. So start a
395+
* new buffer IO here.
383396
384-
/*
397+
*
385398
* wierd race condition:
386399
*
387400
* We were waiting for someone else to read the buffer. While
@@ -396,13 +409,14 @@ BufferAlloc(Relation reln,
396409
*
397410
* This is never going to happen, don't worry about it.
398411
*/
399-
*foundPtr= FALSE;
400-
}
412+
*foundPtr= FALSE;
401413
}
402414
#ifdefBMTRACE
403415
_bm_trace((reln->rd_rel->relisshared ?0 :MyDatabaseId),RelationGetRelid(reln),blockNum,BufferDescriptorGetBuffer(buf),BMT_ALLOCFND);
404416
#endif/* BMTRACE */
405417

418+
if (!(*foundPtr))
419+
StartBufferIO(buf, true);
406420
SpinRelease(BufMgrLock);
407421

408422
returnbuf;
@@ -454,17 +468,15 @@ BufferAlloc(Relation reln,
454468
* in WaitIO until we're done.
455469
*/
456470
inProgress= TRUE;
457-
buf->flags |=BM_IO_IN_PROGRESS;
458471
#ifdefHAS_TEST_AND_SET
459472

460473
/*
461474
* All code paths that acquire this lock pin the buffer first;
462475
* since no one had it pinned (it just came off the free
463476
* list), no one else can have this lock.
464477
*/
465-
Assert(S_LOCK_FREE(&(buf->io_in_progress_lock)));
466-
S_LOCK(&(buf->io_in_progress_lock));
467478
#endif/* HAS_TEST_AND_SET */
479+
StartBufferIO(buf, false);
468480

469481
/*
470482
* Write the buffer out, being careful to release BufMgrLock
@@ -486,12 +498,7 @@ BufferAlloc(Relation reln,
486498
inProgress= FALSE;
487499
buf->flags |=BM_IO_ERROR;
488500
buf->flags &= ~BM_IO_IN_PROGRESS;
489-
#ifdefHAS_TEST_AND_SET
490-
S_UNLOCK(&(buf->io_in_progress_lock));
491-
#else/* !HAS_TEST_AND_SET */
492-
if (buf->refcount>1)
493-
SignalIO(buf);
494-
#endif/* !HAS_TEST_AND_SET */
501+
TerminateBufferIO(buf);
495502
PrivateRefCount[BufferDescriptorGetBuffer(buf)-1]=0;
496503
Assert(buf->refcount>0);
497504
buf->refcount--;
@@ -533,12 +540,7 @@ BufferAlloc(Relation reln,
533540
{
534541
inProgress= FALSE;
535542
buf->flags &= ~BM_IO_IN_PROGRESS;
536-
#ifdefHAS_TEST_AND_SET
537-
S_UNLOCK(&(buf->io_in_progress_lock));
538-
#else/* !HAS_TEST_AND_SET */
539-
if (buf->refcount>1)
540-
SignalIO(buf);
541-
#endif/* !HAS_TEST_AND_SET */
543+
TerminateBufferIO(buf);
542544
PrivateRefCount[BufferDescriptorGetBuffer(buf)-1]=0;
543545
buf->refcount--;
544546
buf= (BufferDesc*)NULL;
@@ -563,12 +565,7 @@ BufferAlloc(Relation reln,
563565
*/
564566
if (buf!=NULL)
565567
{
566-
#ifdefHAS_TEST_AND_SET
567-
S_UNLOCK(&(buf->io_in_progress_lock));
568-
#else/* !HAS_TEST_AND_SET */
569-
if (buf->refcount>1)
570-
SignalIO(buf);
571-
#endif/* !HAS_TEST_AND_SET */
568+
TerminateBufferIO(buf);
572569
/* give up the buffer since we don't need it any more */
573570
PrivateRefCount[BufferDescriptorGetBuffer(buf)-1]=0;
574571
Assert(buf->refcount>0);
@@ -588,10 +585,13 @@ BufferAlloc(Relation reln,
588585
if (inProgress)
589586
{
590587
WaitIO(buf2,BufMgrLock);
591-
if (buf2->flags&BM_IO_ERROR)
592-
*foundPtr= FALSE;
588+
inProgress= (buf2->flags&BM_IO_IN_PROGRESS);
593589
}
590+
if (BUFFER_IS_BROKEN(buf2))
591+
*foundPtr= FALSE;
594592

593+
if (!(*foundPtr))
594+
StartBufferIO(buf2, true);
595595
SpinRelease(BufMgrLock);
596596

597597
returnbuf2;
@@ -640,12 +640,10 @@ BufferAlloc(Relation reln,
640640
*/
641641
if (!inProgress)
642642
{
643-
buf->flags |=BM_IO_IN_PROGRESS;
644-
#ifdefHAS_TEST_AND_SET
645-
Assert(S_LOCK_FREE(&(buf->io_in_progress_lock)));
646-
S_LOCK(&(buf->io_in_progress_lock));
647-
#endif/* HAS_TEST_AND_SET */
643+
StartBufferIO(buf, true);
648644
}
645+
else
646+
ContinueBufferIO(buf, true);
649647

650648
#ifdefBMTRACE
651649
_bm_trace((reln->rd_rel->relisshared ?0 :MyDatabaseId),RelationGetRelid(reln),blockNum,BufferDescriptorGetBuffer(buf),BMT_ALLOCNOTFND);
@@ -806,7 +804,9 @@ FlushBuffer(Buffer buffer, bool release)
806804

807805
/* To check if block content changed while flushing. - vadim 01/17/97 */
808806
SpinAcquire(BufMgrLock);
807+
WaitIO(bufHdr,BufMgrLock);/* confirm end of IO */
809808
bufHdr->flags &= ~BM_JUST_DIRTIED;
809+
StartBufferIO(bufHdr, false);/* output IO start */
810810
SpinRelease(BufMgrLock);
811811

812812
status=smgrflush(DEFAULT_SMGR,bufrel,bufHdr->tag.blockNum,
@@ -824,6 +824,8 @@ FlushBuffer(Buffer buffer, bool release)
824824
BufferFlushCount++;
825825

826826
SpinAcquire(BufMgrLock);
827+
bufHdr->flags &= ~BM_IO_IN_PROGRESS;/* mark IO finished */
828+
TerminateBufferIO(bufHdr);/* output IO finished */
827829

828830
/*
829831
* If this buffer was marked by someone as DIRTY while we were
@@ -998,7 +1000,9 @@ BufferSync()
9981000
* To check if block content changed while flushing (see
9991001
* below). - vadim 01/17/97
10001002
*/
1003+
WaitIO(bufHdr,BufMgrLock);/* confirm end of IO */
10011004
bufHdr->flags &= ~BM_JUST_DIRTIED;
1005+
StartBufferIO(bufHdr, false);/* output IO start */
10021006

10031007
/*
10041008
* If we didn't have the reldesc in our local cache, flush
@@ -1034,6 +1038,8 @@ BufferSync()
10341038
elog(ERROR,"BufferSync: cannot write %u for %s",
10351039
bufHdr->tag.blockNum,bufHdr->sb_relname);
10361040
}
1041+
bufHdr->flags &= ~BM_IO_IN_PROGRESS;/* mark IO finished */
1042+
TerminateBufferIO(bufHdr);/* Sync IO finished */
10371043
BufferFlushCount++;
10381044

10391045
/*
@@ -1084,10 +1090,16 @@ BufferSync()
10841090
staticvoid
10851091
WaitIO(BufferDesc*buf,SPINLOCKspinlock)
10861092
{
1087-
SpinRelease(spinlock);
1088-
S_LOCK(&(buf->io_in_progress_lock));
1089-
S_UNLOCK(&(buf->io_in_progress_lock));
1090-
SpinAcquire(spinlock);
1093+
/*
1094+
* Changed to wait until there's no IO - Inoue 01/13/2000
1095+
*/
1096+
while ((buf->flags&BM_IO_IN_PROGRESS)!=0)
1097+
{
1098+
SpinRelease(spinlock);
1099+
S_LOCK(&(buf->io_in_progress_lock));
1100+
S_UNLOCK(&(buf->io_in_progress_lock));
1101+
SpinAcquire(spinlock);
1102+
}
10911103
}
10921104

10931105
#else/* !HAS_TEST_AND_SET */
@@ -2163,3 +2175,112 @@ LockBuffer(Buffer buffer, int mode)
21632175
#endif
21642176

21652177
}
2178+
2179+
/*
2180+
*Functions for IO error handling
2181+
*
2182+
*Note : We assume that nested buffer IO never occur.
2183+
*i.e at most one io_in_progress spinlock is held
2184+
*per proc.
2185+
*/
2186+
staticBufferDesc*InProgressBuf= (BufferDesc*)NULL;
2187+
staticboolIsForInput;
2188+
2189+
/*
2190+
* Function:StartBufferIO
2191+
*(Assumptions)
2192+
*My process is executing no IO
2193+
*BufMgrLock is held
2194+
*BM_IO_IN_PROGRESS mask is not set for the buffer
2195+
*The buffer is Pinned
2196+
*
2197+
*/
2198+
staticvoidStartBufferIO(BufferDesc*buf,boolforInput)
2199+
{
2200+
Assert(!InProgressBuf);
2201+
Assert(!(buf->flags&BM_IO_IN_PROGRESS));
2202+
buf->flags |=BM_IO_IN_PROGRESS;
2203+
#ifdefHAS_TEST_AND_SET
2204+
Assert(S_LOCK_FREE(&(buf->io_in_progress_lock)))
2205+
S_LOCK(&(buf->io_in_progress_lock));
2206+
#endif/* HAS_TEST_AND_SET */
2207+
InProgressBuf=buf;
2208+
IsForInput=forInput;
2209+
}
2210+
2211+
/*
2212+
* Function:TerminateBufferIO
2213+
*(Assumptions)
2214+
*My process is executing IO for the buffer
2215+
*BufMgrLock is held
2216+
*The buffer is Pinned
2217+
*
2218+
*/
2219+
staticvoidTerminateBufferIO(BufferDesc*buf)
2220+
{
2221+
Assert(buf==InProgressBuf);
2222+
#ifdefHAS_TEST_AND_SET
2223+
S_UNLOCK(&(buf->io_in_progress_lock));
2224+
#else
2225+
if (buf->refcount>1)
2226+
SignalIO(buf);
2227+
#endif/* HAS_TEST_AND_SET */
2228+
InProgressBuf= (BufferDesc*)0;
2229+
}
2230+
2231+
/*
2232+
* Function:ContinueBufferIO
2233+
*(Assumptions)
2234+
*My process is executing IO for the buffer
2235+
*BufMgrLock is held
2236+
*The buffer is Pinned
2237+
*
2238+
*/
2239+
staticvoidContinueBufferIO(BufferDesc*buf,boolforInput)
2240+
{
2241+
Assert(buf==InProgressBuf);
2242+
Assert(buf->flags&BM_IO_IN_PROGRESS);
2243+
IsForInput=forInput;
2244+
}
2245+
2246+
externvoidInitBufferIO(void)
2247+
{
2248+
InProgressBuf= (BufferDesc*)0;
2249+
}
2250+
2251+
/*
2252+
*This function is called from ProcReleaseSpins().
2253+
*BufMgrLock isn't held when this function is called.
2254+
*BM_IO_ERROR is always set. If BM_IO_ERROR was already
2255+
*set in case of output,this routine would kill all
2256+
*backends and reset postmaster.
2257+
*/
2258+
externvoidAbortBufferIO(void)
2259+
{
2260+
BufferDesc*buf=InProgressBuf;
2261+
if (buf)
2262+
{
2263+
Assert(buf->flags&BM_IO_IN_PROGRESS);
2264+
SpinAcquire(BufMgrLock);
2265+
if (IsForInput)
2266+
{
2267+
Assert(!(buf->flags&BM_DIRTY));
2268+
}
2269+
else
2270+
{
2271+
Assert(!(buf->flags&BM_DIRTY));
2272+
/* Assert(!(buf->flags & BM_IO_ERROR)); */
2273+
if (buf->flags&BM_IO_ERROR)
2274+
{
2275+
elog(NOTICE,"!!! write error seems permanent !!!");
2276+
elog(NOTICE,"!!! now kill all backends and reset postmaster !!!");
2277+
proc_exit(255);
2278+
}
2279+
buf->flags |=BM_DIRTY;
2280+
}
2281+
buf->flags |=BM_IO_ERROR;
2282+
TerminateBufferIO(buf);
2283+
buf->flags &= ~BM_IO_IN_PROGRESS;
2284+
SpinRelease(BufMgrLock);
2285+
}
2286+
}

‎src/backend/storage/lmgr/proc.c

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -7,7 +7,7 @@
77
*
88
*
99
* IDENTIFICATION
10-
* $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.65 1999/12/16 01:25:08 momjian Exp $
10+
* $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.66 2000/01/17 01:15:18 inoue Exp $
1111
*
1212
*-------------------------------------------------------------------------
1313
*/
@@ -46,7 +46,7 @@
4646
*This is so that we can support more backends. (system-wide semaphore
4747
*sets run out pretty fast.) -ay 4/95
4848
*
49-
* $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.65 1999/12/16 01:25:08 momjian Exp $
49+
* $Header: /cvsroot/pgsql/src/backend/storage/lmgr/proc.c,v 1.66 2000/01/17 01:15:18 inoue Exp $
5050
*/
5151
#include<sys/time.h>
5252
#include<unistd.h>
@@ -848,6 +848,7 @@ ProcReleaseSpins(PROC *proc)
848848
SpinRelease(i);
849849
}
850850
}
851+
AbortBufferIO();
851852
}
852853

853854
/*****************************************************************************

‎src/include/storage/bufmgr.h

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,7 +6,7 @@
66
*
77
* Copyright (c) 1994, Regents of the University of California
88
*
9-
* $Id: bufmgr.h,v 1.32 1999/09/28 11:40:53 vadim Exp $
9+
* $Id: bufmgr.h,v 1.33 2000/01/17 01:15:19 inoue Exp $
1010
*
1111
*-------------------------------------------------------------------------
1212
*/
@@ -186,5 +186,6 @@ extern void SetBufferCommitInfoNeedsSave(Buffer buffer);
186186

187187
externvoidUnlockBuffers(void);
188188
externvoidLockBuffer(Bufferbuffer,intmode);
189+
externvoidAbortBufferIO(void);
189190

190191
#endif

0 commit comments

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