Re: make the stats collector shutdown without writing the statsfiles if the immediate shutdown is requested.

From: Fujii Masao <masao(dot)fujii(at)oss(dot)nttdata(dot)com>
To: Masahiro Ikeda <ikedamsh(at)oss(dot)nttdata(dot)com>
Cc: kuroda(dot)hayato(at)fujitsu(dot)com, Zhihong Yu <zyu(at)yugabyte(dot)com>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: make the stats collector shutdown without writing the statsfiles if the immediate shutdown is requested.
Date: 2021-03-23 02:40:26
Message-ID: ae2a2c43-d530-7c33-c510-544d61ea76b1@oss.nttdata.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

On 2021/03/23 9:05, Masahiro Ikeda wrote:
> Yes. I attached the v5 patch based on v3 patch.
> I renamed SignalHandlerForUnsafeExit() and fixed the following comment.

Thanks for updating the patch!

When the startup process exits because of recovery_target_action=shutdown,
reaper() calls TerminateChildren(SIGTERM). This function sends SIGTERM to
the stats collector. Currently the stats collector ignores SIGTERM, but with
the patch it exits normally. This change of behavior might be problematic.

That is, TerminateChildren(SIGTERM) sends SIGTERM to various processes.
But currently the stats collector and checkpointer don't exit even when
SIGTERM arrives because they ignore SIGTERM. After several processes
other than the stats collector and checkpointer exit by SIGTERM,
PostmasterStateMachine() and reaper() make checkpointer exit and then
the stats collector exit. The shutdown terminates the processes in this order.

On the other hand, with the patch, the stats collector exits by SIGTERM
before checkpointer exits. This is not normal order of processes to exit in
shutdown.

To address this issue, one idea is to use SIGUSR2 for normal exit of the stats
collector, instead of SIGTERM. If we do this, TerminateChildren(SIGTERM)
cannot terminate the stats collector. Thought?

If we adopt this idea, the detail comment about why SIGUSR2 is used for that
needs to be added.

Regards,

--
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message tsunakawa.takay@fujitsu.com 2021-03-23 03:11:25 RE: Disable WAL logging to speed up data loading
Previous Message Kyotaro Horiguchi 2021-03-23 02:20:52 Re: shared-memory based stats collector