Horton

Re: Weird failure with latches in curculio on v15

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andres Freund <andres@anarazel.de>
Cc: Thomas Munro <thomas.munro@gmail.com>, Fujii Masao <fujii@postgresql.org>, Nathan Bossart <nathandbossart@gmail.com>, Michael Paquier <michael@paquier.xyz>, Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2023-02-01T15:12:26Z
Lists: pgsql-hackers 
Andres Freund <andres@anarazel.de> writes:
> On 2023-02-01 16:21:16 +1300, Thomas Munro wrote:
>> It's always in proc_exit() in StartupProcShutdownHandler(), a SIGTERM
>> handler which is allowed to call that while in_restore_command is
>> true.

> Ugh, no wonder we're getting crashes. This whole business seems bogus as
> hell.

Indeed :-(

> I don't see a choice but to revert the recent changes. They need a
> fairly large rewrite.

9a740f81e clearly made things a lot worse, but it wasn't great
before.  Can we see a way forward to removing the problem entirely?

The fundamental issue is that we have no good way to break out
of system(), and I think the original idea was that
in_restore_command would be set *only* for the duration of the
system() call.  That's clearly been lost sight of completely,
but maybe as a stopgap we could try to get back to that.

			regards, tom lane

Commits

  1. Avoid calling proc_exit() in processes forked by system().

  2. Move extra code out of the Pre/PostRestoreCommand() section.

  3. Revert refactoring of restore command code to shell_restore.c

  4. Refactor code in charge of running shell-based recovery commands

  5. Clean up inconsistent use of fflush().

  6. Report wait events for local shell commands like archive_command.