Re: pg_terminate_backend() issues

Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > Tom Lane wrote:
> >> I was able to get things to more or less work most of the time with
> >> three or four additional ugly hacks in postgres.c, but I still don't
> >> have any great confidence that there aren't windows where a terminate
> >> request wouldn't be ignored (even without considering the uncooperative-
> >> function scenario). Moreover it's entirely unclear that this approach
> >> actually dodges any of the hypothetical bugs in SIGTERM handling.
>
> > I don't understand. If we call proc_exit(0) instead, it is the same as
> > someone exiting the backend via PQfinish().
>
> Well, using proc_exit() instead of die() might have alleviated that
> particular objection, but there are still the others.

Tom, thanks for you feedback. It was very helpful.

I have adjusted the patch to return a locked PGPROC entry, initialize
the PGPROC->terminate boolean, called proc_exit(), and moved the
PGPROC->terminate test out of the cancel loop entirely so lost cancel
signals are not as big a problem anymore.

I agree it would be best for pg_terminate_backend() and for the
postmaster use of SIGTERM if SIGTERM was more reliable about cleanup, so
hopefully that will work for 8.4. I have attached my patch in hopes we
can use it as a backup in case we can't get SIGTERM to work for
individual backends for 8.4.

[ I have posted about testing in a separate email.]

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ If your life is a hard drive, Christ can be your backup. +