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

Commit974ece5

Browse files
committed
Fix an assertion failure related to an exclusive backup.
Previously multiple sessions could execute pg_start_backup() andpg_stop_backup() to start and stop an exclusive backup at the same time.This could trigger the assertion failure of"FailedAssertion("!(XLogCtl->Insert.exclusiveBackup)".This happend because, even while pg_start_backup() was startingan exclusive backup, other session could run pg_stop_backup()concurrently and mark the backup as not-in-progress unconditionally.This patch introduces ExclusiveBackupState indicating the state ofan exclusive backup. This state is used to ensure that there is onlyone session running pg_start_backup() or pg_stop_backup() atthe same time, to avoid the assertion failure.Back-patch to all supported versions.Author: Michael PaquierReviewed-By: Kyotaro Horiguchi and meReported-By: Andreas SeltenreichDiscussion: <87mvktojme.fsf@credativ.de>
1 parentd43a619 commit974ece5

File tree

1 file changed

+150
-73
lines changed
  • src/backend/access/transam

1 file changed

+150
-73
lines changed

‎src/backend/access/transam/xlog.c

Lines changed: 150 additions & 73 deletions
Original file line numberDiff line numberDiff line change
@@ -472,6 +472,29 @@ typedef union WALInsertLockPadded
472472
charpad[PG_CACHE_LINE_SIZE];
473473
}WALInsertLockPadded;
474474

475+
/*
476+
* State of an exclusive backup, necessary to control concurrent activities
477+
* across sessions when working on exclusive backups.
478+
*
479+
* EXCLUSIVE_BACKUP_NONE means that there is no exclusive backup actually
480+
* running, to be more precise pg_start_backup() is not being executed for
481+
* an exclusive backup and there is no exclusive backup in progress.
482+
* EXCLUSIVE_BACKUP_STARTING means that pg_start_backup() is starting an
483+
* exclusive backup.
484+
* EXCLUSIVE_BACKUP_IN_PROGRESS means that pg_start_backup() has finished
485+
* running and an exclusive backup is in progress. pg_stop_backup() is
486+
* needed to finish it.
487+
* EXCLUSIVE_BACKUP_STOPPING means that pg_stop_backup() is stopping an
488+
* exclusive backup.
489+
*/
490+
typedefenumExclusiveBackupState
491+
{
492+
EXCLUSIVE_BACKUP_NONE=0,
493+
EXCLUSIVE_BACKUP_STARTING,
494+
EXCLUSIVE_BACKUP_IN_PROGRESS,
495+
EXCLUSIVE_BACKUP_STOPPING
496+
}ExclusiveBackupState;
497+
475498
/*
476499
* Shared state data for WAL insertion.
477500
*/
@@ -513,13 +536,15 @@ typedef struct XLogCtlInsert
513536
boolfullPageWrites;
514537

515538
/*
516-
* exclusiveBackup is true if a backup started with pg_start_backup() is
517-
* in progress, and nonExclusiveBackups is a counter indicating the number
518-
* of streaming base backups currently in progress. forcePageWrites is set
519-
* to true when either of these is non-zero. lastBackupStart is the latest
520-
* checkpoint redo location used as a starting point for an online backup.
539+
* exclusiveBackupState indicates the state of an exclusive backup
540+
* (see comments of ExclusiveBackupState for more details).
541+
* nonExclusiveBackups is a counter indicating the number of streaming
542+
* base backups currently in progress. forcePageWrites is set to true
543+
* when either of these is non-zero. lastBackupStart is the latest
544+
* checkpoint redo location used as a starting point for an online
545+
* backup.
521546
*/
522-
boolexclusiveBackup;
547+
ExclusiveBackupStateexclusiveBackupState;
523548
intnonExclusiveBackups;
524549
XLogRecPtrlastBackupStart;
525550

@@ -858,6 +883,7 @@ static void xlog_outrec(StringInfo buf, XLogReaderState *record);
858883
#endif
859884
staticvoidxlog_outdesc(StringInfobuf,XLogReaderState*record);
860885
staticvoidpg_start_backup_callback(intcode,Datumarg);
886+
staticvoidpg_stop_backup_callback(intcode,Datumarg);
861887
staticboolread_backup_label(XLogRecPtr*checkPointLoc,
862888
bool*backupEndRequired,bool*backupFromStandby);
863889
staticboolread_tablespace_map(List**tablespaces);
@@ -10016,15 +10042,20 @@ do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
1001610042
WALInsertLockAcquireExclusive();
1001710043
if (exclusive)
1001810044
{
10019-
if (XLogCtl->Insert.exclusiveBackup)
10045+
/*
10046+
* At first, mark that we're now starting an exclusive backup,
10047+
* to ensure that there are no other sessions currently running
10048+
* pg_start_backup() or pg_stop_backup().
10049+
*/
10050+
if (XLogCtl->Insert.exclusiveBackupState!=EXCLUSIVE_BACKUP_NONE)
1002010051
{
1002110052
WALInsertLockRelease();
1002210053
ereport(ERROR,
1002310054
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1002410055
errmsg("a backup is already in progress"),
1002510056
errhint("Run pg_stop_backup() and try again.")));
1002610057
}
10027-
XLogCtl->Insert.exclusiveBackup=true;
10058+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_STARTING;
1002810059
}
1002910060
else
1003010061
XLogCtl->Insert.nonExclusiveBackups++;
@@ -10279,7 +10310,7 @@ do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
1027910310
{
1028010311
/*
1028110312
* Check for existing backup label --- implies a backup is already
10282-
* running. (XXX given that we checkedexclusiveBackup above,
10313+
* running. (XXX given that we checkedexclusiveBackupState above,
1028310314
* maybe it would be OK to just unlink any such label file?)
1028410315
*/
1028510316
if (stat(BACKUP_LABEL_FILE,&stat_buf)!=0)
@@ -10360,6 +10391,16 @@ do_pg_start_backup(const char *backupidstr, bool fast, TimeLineID *starttli_p,
1036010391
}
1036110392
PG_END_ENSURE_ERROR_CLEANUP(pg_start_backup_callback, (Datum)BoolGetDatum(exclusive));
1036210393

10394+
/*
10395+
* Mark that start phase has correctly finished for an exclusive backup.
10396+
*/
10397+
if (exclusive)
10398+
{
10399+
WALInsertLockAcquireExclusive();
10400+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_IN_PROGRESS;
10401+
WALInsertLockRelease();
10402+
}
10403+
1036310404
/*
1036410405
* We're done. As a convenience, return the starting WAL location.
1036510406
*/
@@ -10378,23 +10419,41 @@ pg_start_backup_callback(int code, Datum arg)
1037810419
WALInsertLockAcquireExclusive();
1037910420
if (exclusive)
1038010421
{
10381-
Assert(XLogCtl->Insert.exclusiveBackup);
10382-
XLogCtl->Insert.exclusiveBackup=false;
10422+
Assert(XLogCtl->Insert.exclusiveBackupState==EXCLUSIVE_BACKUP_STARTING);
10423+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_NONE;
1038310424
}
1038410425
else
1038510426
{
1038610427
Assert(XLogCtl->Insert.nonExclusiveBackups>0);
1038710428
XLogCtl->Insert.nonExclusiveBackups--;
1038810429
}
1038910430

10390-
if (!XLogCtl->Insert.exclusiveBackup&&
10431+
if (XLogCtl->Insert.exclusiveBackupState==EXCLUSIVE_BACKUP_NONE&&
1039110432
XLogCtl->Insert.nonExclusiveBackups==0)
1039210433
{
1039310434
XLogCtl->Insert.forcePageWrites= false;
1039410435
}
1039510436
WALInsertLockRelease();
1039610437
}
1039710438

10439+
/*
10440+
* Error cleanup callback for pg_stop_backup
10441+
*/
10442+
staticvoid
10443+
pg_stop_backup_callback(intcode,Datumarg)
10444+
{
10445+
boolexclusive=DatumGetBool(arg);
10446+
10447+
/* Update backup status on failure */
10448+
WALInsertLockAcquireExclusive();
10449+
if (exclusive)
10450+
{
10451+
Assert(XLogCtl->Insert.exclusiveBackupState==EXCLUSIVE_BACKUP_STOPPING);
10452+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_IN_PROGRESS;
10453+
}
10454+
WALInsertLockRelease();
10455+
}
10456+
1039810457
/*
1039910458
* do_pg_stop_backup is the workhorse of the user-visible pg_stop_backup()
1040010459
* function.
@@ -10457,20 +10516,91 @@ do_pg_stop_backup(char *labelfile, bool waitforarchive, TimeLineID *stoptli_p)
1045710516
errmsg("WAL level not sufficient for making an online backup"),
1045810517
errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
1045910518

10460-
/*
10461-
* OK to update backup counters and forcePageWrites
10462-
*/
10463-
WALInsertLockAcquireExclusive();
1046410519
if (exclusive)
1046510520
{
10466-
if (!XLogCtl->Insert.exclusiveBackup)
10521+
/*
10522+
* At first, mark that we're now stopping an exclusive backup,
10523+
* to ensure that there are no other sessions currently running
10524+
* pg_start_backup() or pg_stop_backup().
10525+
*/
10526+
WALInsertLockAcquireExclusive();
10527+
if (XLogCtl->Insert.exclusiveBackupState!=EXCLUSIVE_BACKUP_IN_PROGRESS)
1046710528
{
1046810529
WALInsertLockRelease();
1046910530
ereport(ERROR,
1047010531
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1047110532
errmsg("exclusive backup not in progress")));
1047210533
}
10473-
XLogCtl->Insert.exclusiveBackup= false;
10534+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_STOPPING;
10535+
WALInsertLockRelease();
10536+
10537+
/*
10538+
* Remove backup_label. In case of failure, the state for an exclusive
10539+
* backup is switched back to in-progress.
10540+
*/
10541+
PG_ENSURE_ERROR_CLEANUP(pg_stop_backup_callback, (Datum)BoolGetDatum(exclusive));
10542+
{
10543+
/*
10544+
* Read the existing label file into memory.
10545+
*/
10546+
structstatstatbuf;
10547+
intr;
10548+
10549+
if (stat(BACKUP_LABEL_FILE,&statbuf))
10550+
{
10551+
/* should not happen per the upper checks */
10552+
if (errno!=ENOENT)
10553+
ereport(ERROR,
10554+
(errcode_for_file_access(),
10555+
errmsg("could not stat file \"%s\": %m",
10556+
BACKUP_LABEL_FILE)));
10557+
ereport(ERROR,
10558+
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10559+
errmsg("a backup is not in progress")));
10560+
}
10561+
10562+
lfp=AllocateFile(BACKUP_LABEL_FILE,"r");
10563+
if (!lfp)
10564+
{
10565+
ereport(ERROR,
10566+
(errcode_for_file_access(),
10567+
errmsg("could not read file \"%s\": %m",
10568+
BACKUP_LABEL_FILE)));
10569+
}
10570+
labelfile=palloc(statbuf.st_size+1);
10571+
r=fread(labelfile,statbuf.st_size,1,lfp);
10572+
labelfile[statbuf.st_size]='\0';
10573+
10574+
/*
10575+
* Close and remove the backup label file
10576+
*/
10577+
if (r!=1||ferror(lfp)||FreeFile(lfp))
10578+
ereport(ERROR,
10579+
(errcode_for_file_access(),
10580+
errmsg("could not read file \"%s\": %m",
10581+
BACKUP_LABEL_FILE)));
10582+
if (unlink(BACKUP_LABEL_FILE)!=0)
10583+
ereport(ERROR,
10584+
(errcode_for_file_access(),
10585+
errmsg("could not remove file \"%s\": %m",
10586+
BACKUP_LABEL_FILE)));
10587+
10588+
/*
10589+
* Remove tablespace_map file if present, it is created only if there
10590+
* are tablespaces.
10591+
*/
10592+
unlink(TABLESPACE_MAP);
10593+
}
10594+
PG_END_ENSURE_ERROR_CLEANUP(pg_stop_backup_callback, (Datum)BoolGetDatum(exclusive));
10595+
}
10596+
10597+
/*
10598+
* OK to update backup counters and forcePageWrites
10599+
*/
10600+
WALInsertLockAcquireExclusive();
10601+
if (exclusive)
10602+
{
10603+
XLogCtl->Insert.exclusiveBackupState=EXCLUSIVE_BACKUP_NONE;
1047410604
}
1047510605
else
1047610606
{
@@ -10484,66 +10614,13 @@ do_pg_stop_backup(char *labelfile, bool waitforarchive, TimeLineID *stoptli_p)
1048410614
XLogCtl->Insert.nonExclusiveBackups--;
1048510615
}
1048610616

10487-
if (!XLogCtl->Insert.exclusiveBackup&&
10617+
if (XLogCtl->Insert.exclusiveBackupState==EXCLUSIVE_BACKUP_NONE&&
1048810618
XLogCtl->Insert.nonExclusiveBackups==0)
1048910619
{
1049010620
XLogCtl->Insert.forcePageWrites= false;
1049110621
}
1049210622
WALInsertLockRelease();
1049310623

10494-
if (exclusive)
10495-
{
10496-
/*
10497-
* Read the existing label file into memory.
10498-
*/
10499-
structstatstatbuf;
10500-
intr;
10501-
10502-
if (stat(BACKUP_LABEL_FILE,&statbuf))
10503-
{
10504-
if (errno!=ENOENT)
10505-
ereport(ERROR,
10506-
(errcode_for_file_access(),
10507-
errmsg("could not stat file \"%s\": %m",
10508-
BACKUP_LABEL_FILE)));
10509-
ereport(ERROR,
10510-
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
10511-
errmsg("a backup is not in progress")));
10512-
}
10513-
10514-
lfp=AllocateFile(BACKUP_LABEL_FILE,"r");
10515-
if (!lfp)
10516-
{
10517-
ereport(ERROR,
10518-
(errcode_for_file_access(),
10519-
errmsg("could not read file \"%s\": %m",
10520-
BACKUP_LABEL_FILE)));
10521-
}
10522-
labelfile=palloc(statbuf.st_size+1);
10523-
r=fread(labelfile,statbuf.st_size,1,lfp);
10524-
labelfile[statbuf.st_size]='\0';
10525-
10526-
/*
10527-
* Close and remove the backup label file
10528-
*/
10529-
if (r!=1||ferror(lfp)||FreeFile(lfp))
10530-
ereport(ERROR,
10531-
(errcode_for_file_access(),
10532-
errmsg("could not read file \"%s\": %m",
10533-
BACKUP_LABEL_FILE)));
10534-
if (unlink(BACKUP_LABEL_FILE)!=0)
10535-
ereport(ERROR,
10536-
(errcode_for_file_access(),
10537-
errmsg("could not remove file \"%s\": %m",
10538-
BACKUP_LABEL_FILE)));
10539-
10540-
/*
10541-
* Remove tablespace_map file if present, it is created only if there
10542-
* are tablespaces.
10543-
*/
10544-
unlink(TABLESPACE_MAP);
10545-
}
10546-
1054710624
/*
1054810625
* Read and parse the START WAL LOCATION line (this code is pretty crude,
1054910626
* but we are not expecting any variability in the file format).
@@ -10780,7 +10857,7 @@ do_pg_abort_backup(void)
1078010857
Assert(XLogCtl->Insert.nonExclusiveBackups>0);
1078110858
XLogCtl->Insert.nonExclusiveBackups--;
1078210859

10783-
if (!XLogCtl->Insert.exclusiveBackup&&
10860+
if (XLogCtl->Insert.exclusiveBackupState==EXCLUSIVE_BACKUP_NONE&&
1078410861
XLogCtl->Insert.nonExclusiveBackups==0)
1078510862
{
1078610863
XLogCtl->Insert.forcePageWrites= false;

0 commit comments

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