Horton

Re: BUG #15804: Assertion failure when using logging_collector with EXEC_BACKEND

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Yuli Khodorkovskiy <yuli.khodorkovskiy@crunchydata.com>, pgsql-bugs@lists.postgresql.org, Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Date: 2019-05-16T04:36:18Z
Lists: pgsql-bugs 
On Thu, May 16, 2019 at 12:56:17PM +0900, Michael Paquier wrote:
> The test is able to pass, but we have a race condition between the
> moment the backend file gets saved and the moment we allow it to be
> read.   I have not spent much time checking the stack between
> InitializeMaxBackends() and RemovePgTempFiles() in postmaster.c, but
> 57431a9 triggers the failure.

Oh, I think I got it.  The issue is that we call RemovePgTempFiles()
after starting the syslogger.  This cannot be run with other processes
running in parallel, and with EXEC_BACKEND there is the extra case
where we have a pgsql_tmp/ at the root of the data folder, so the
syslogger complains on that.  By making RemovePgTempFiles() happen
before starting the syslogger, then the test complains again about the
assertion without my previous patch applied of course.  With the patch
applied, I get no complains.
--
Michael

Commits

  1. Rearrange postmaster's startup sequence for better syslogger results.

  2. Improve logrotate test so that it meaningfully exercises syslogger.

  3. Revert "postmaster: Start syslogger earlier".

  4. postmaster: Start syslogger earlier