Re: processes stuck in shutdown following OOM/recovery

From: Thomas Munro <thomas(dot)munro(at)gmail(dot)com>
To: Justin Pryzby <pryzby(at)telsasoft(dot)com>
Cc: pgsql-hackers(at)postgresql(dot)org, Thomas Munro <tmunro(at)postgresql(dot)org>
Subject: Re: processes stuck in shutdown following OOM/recovery
Date: 2024-03-05 21:22:09
Message-ID: CA+hUKGKOW1ON3Cae1OUtHEPyFKc7O_XGGrUKfa9t5Ce0aLbeFQ@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On Sat, Dec 2, 2023 at 3:30 PM Thomas Munro <thomas(dot)munro(at)gmail(dot)com> wrote:
> On Sat, Dec 2, 2023 at 2:18 PM Thomas Munro <thomas(dot)munro(at)gmail(dot)com> wrote:
> > On Fri, Dec 1, 2023 at 6:13 PM Justin Pryzby <pryzby(at)telsasoft(dot)com> wrote:
> > > $ kill -9 2524495; sleep 0.05; pg_ctl -D ./pgdev.dat1 stop -m fast # 2524495 is a child's pid
> >
> > > This affects v15, and fails at ) but not its parent.
> >
> > Repro'd here. I had to make the sleep shorter on my system. Looking...
>
> The PostmasterStateMachine() case for PM_WAIT_BACKENDS doesn't tell
> the checkpointer to shut down in this race case. We have
> CheckpointerPID != 0 (because 7ff23c6d27 starts it earlier than
> before), and FatalError is true because a child recently crashed and
> we haven't yet received the PMSIGNAL_RECOVERY_STARTED handler that
> would clear it. Hmm.

Here is a first attempt at fixing this. I am not yet 100% sure if it
is right, and there may be a nicer/simpler way to express the
conditions. It passes the test suite, and it fixes the repro that
Justin posted. FYI on my machine I had to use sleep 0.005 where he
had 0.05, as an FYI if someone else is trying to reproduce the issue.

Attachment Content-Type Size
0001-Fix-rare-recovery-shutdown-hang-due-to-checkpointer.patch application/octet-stream 3.0 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tomas Vondra 2024-03-05 21:30:20 Re: remaining sql/json patches
Previous Message Alena Rybakina 2024-03-05 20:10:00 Re: Support "Right Semi Join" plan shapes