Re: Race conditions with checkpointer and shutdown

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

From: Tom Lane <tgl@sss.pgh.pa.us>
To: Alvaro Herrera <alvherre@2ndquadrant.com>
Cc: Ashwin Agrawal <aagrawal@pivotal.io>, Michael Paquier <michael@paquier.xyz>, Thomas Munro <thomas.munro@gmail.com>, Postgres hackers <pgsql-hackers@lists.postgresql.org>
Date: 2019-06-12T20:26:23Z
Lists: pgsql-hackers
Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Jun-12, Tom Lane wrote:
>> Should we do that now, or wait till after next week's releases?

> IMO this has been hammered enough in master, and we still have a few
> days in the back-branches for buildfarm, that it's okay to do it now.

Poking at that, I find that a1a789eb5 back-patches reasonably painlessly
into v11 and v10, but trying to bring it back to 9.6 encounters a pile of
merge failures.  Also, looking at the git logs shows that we did a hell
of a lot of subtle work on that code (libpqwalreceiver.c in particular)
during the v10 cycle.  So I've got no confidence that successful
buildfarm/beta1 testing of the HEAD patch means much of anything for
putting it into pre-v10 branches.

Given that we've seen few if any field reports of this issue, my
inclination is to back-patch as far as v10, but not take the risk
and effort involved in going further.

			regards, tom lane



Commits

  1. In walreceiver, don't try to do ereport() in a signal handler.