Horton

Re: BUG #16161: pg_ctl stop fails sometimes (on Windows)

Alexander Law <exclusion@gmail.com>

View thread View on postgr.es
From: Alexander Lakhin <exclusion@gmail.com>
To: pgsql-bugs@lists.postgresql.org, PG Bug reporting form <noreply@postgresql.org>
Date: 2019-12-11T20:30:01Z
Lists: pgsql-bugs

Attachments

11.12.2019 23:00, PG Bug reporting form wrote:
> The following bug has been logged on the website:
>
> Bug reference:      16161
> Logged by:          Alexander Lakhin
> Email address:      exclusion@gmail.com
> PostgreSQL version: 12.1
> Operating system:   Windows
> Description:        
>
> The regression tests on Windows sometimes fail with 'Permission denied'
> errors. For example:
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=dory&dt=2019-12-11%2007%3A45%3A33
>
> ============== shutting down postmaster               ==============
> pg_ctl: could not open PID file
> "c:/pgbuildfarm/pgbuildroot/HEAD/pgsql.build/src/test/regress/./tmp_check/data/postmaster.pid":
> Permission denied
>
> This error occurs when pg_ctl is trying to open postmaster.pid while this
> file is in "delete pending" state
> (https://stackoverflow.com/questions/3764072/).
To reproduce the issue reliably I propose a simple modification to
synchronize unlink() with open() and a simple test
(sync_pid_ops+test.patch).
With the patch applied, `vcregress taptest src/test/restart` fails for
me on iteration 47, 6, 7, 42, 51, 26, ...

I see two ways to fix the issue:
1) Unlink postmaster.pid using rename operation (adopt the solution from
https://stackoverflow.com/questions/3764072/).
2) Ignore such 'Permission denied' error and just try to open the file
once again (attached fix_open_for_unlink.patch implements this).


I'm inclined to the second approach as pgwin32_open() already handles
two transient states (Windows-only), and it could be useful not only for
postmaster.pid, but for some other files.


Best regards,
Alexander

Commits

  1. In pgwin32_open, loop after ERROR_ACCESS_DENIED only if we can't stat.

  2. On Windows, wait a little to see if ERROR_ACCESS_DENIED goes away.