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

Commit8ff5186

Browse files
committed
Fix confusion of max_parallel_workers mechanism following crash.
Commitb460f5d failed to contemplatethe possibilit that a parallel worker registered before a crash wouldbe unregistered only after the crash; if that happened, we'd end upwith parallel_terminate_count > parallel_register_count and thesystem would refuse to launch any more parallel workers.The easiest way to fix that seems to be to forget BGW_NEVER_RESTARTworkers in ResetBackgroundWorkerCrashTimes() rather than leaving themaround to be cleaned up after the conclusion of the restart, so thatthey go away before rather than after shared memory is reset.To make sure that this fix is water-tight, don't allow parallelworkers to be anything other than BGW_NEVER_RESTART, so that afterrecovering from a crash, 0 is guaranteed to be the correct startingvalue for parallel_register_count. The core code wouldn't do thisanyway, but somebody might try to do it in extension code.Report by Thomas Vondra. Patch by me, reviewed by Kuntal Ghosh.Discussion:http://postgr.es/m/CAGz5QC+AVEVS+3rBKRq83AxkJLMZ1peMt4nnrQwczxOrmo3CNw@mail.gmail.com
1 parent1e298b8 commit8ff5186

File tree

1 file changed

+42
-6
lines changed

1 file changed

+42
-6
lines changed

‎src/backend/postmaster/bgworker.c

Lines changed: 42 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -515,13 +515,34 @@ ResetBackgroundWorkerCrashTimes(void)
515515

516516
rw=slist_container(RegisteredBgWorker,rw_lnode,iter.cur);
517517

518-
/*
519-
* For workers that should not be restarted, we don't want to lose the
520-
* information that they have crashed; otherwise, they would be
521-
* restarted, which is wrong.
522-
*/
523-
if (rw->rw_worker.bgw_restart_time!=BGW_NEVER_RESTART)
518+
if (rw->rw_worker.bgw_restart_time==BGW_NEVER_RESTART)
519+
{
520+
/*
521+
* Workers marked BGW_NVER_RESTART shouldn't get relaunched after
522+
* the crash, so forget about them. (If we wait until after the
523+
* crash to forget about them, and they are parallel workers,
524+
* parallel_terminate_count will get incremented after we've
525+
* already zeroed parallel_register_count, which would be bad.)
526+
*/
527+
ForgetBackgroundWorker(&iter);
528+
}
529+
else
530+
{
531+
/*
532+
* The accounting which we do via parallel_register_count and
533+
* parallel_terminate_count would get messed up if a worker marked
534+
* parallel could survive a crash and restart cycle. All such
535+
* workers should be marked BGW_NEVER_RESTART, and thus control
536+
* should never reach this branch.
537+
*/
538+
Assert((rw->rw_worker.bgw_flags&BGWORKER_CLASS_PARALLEL)==0);
539+
540+
/*
541+
* Allow this worker to be restarted immediately after we finish
542+
* resetting.
543+
*/
524544
rw->rw_crashed_at=0;
545+
}
525546
}
526547
}
527548

@@ -589,6 +610,21 @@ SanityCheckBackgroundWorker(BackgroundWorker *worker, int elevel)
589610
return false;
590611
}
591612

613+
/*
614+
* Parallel workers may not be configured for restart, because the
615+
* parallel_register_count/parallel_terminate_count accounting can't
616+
* handle parallel workers lasting through a crash-and-restart cycle.
617+
*/
618+
if (worker->bgw_restart_time!=BGW_NEVER_RESTART&&
619+
(worker->bgw_flags&BGWORKER_CLASS_PARALLEL)!=0)
620+
{
621+
ereport(elevel,
622+
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
623+
errmsg("background worker \"%s\": parallel workers may not be configured for restart",
624+
worker->bgw_name)));
625+
return false;
626+
}
627+
592628
return true;
593629
}
594630

0 commit comments

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