Chris Campbell <chris@bignerdranch.com> writes:
> On Oct 26, 2006, at 17:21, Tom Lane wrote:
>> And what was 1171 doing? I really doubt that either of these could
>> have been pg_dump.
> I know that process 1120 is a Java client (Hibernate) running an
> UPDATE query, but I have no idea what 1171 is. I doubt that 1171 was
> pg_dump, but when we turn off the pg_dump cron jobs (for 12-ish
> hours), the deadlocks go away. We usually see 5 or 6 deadlocks spread
> throughout the day. That's not definitive evidence, of course, but
> it's certainly curious.
FWIW, we recently found a deadlock risk that's likely to be triggered in
situations involving multiple updates and foreign keys. I still don't
see any connection to pg_dump, but otherwise it seems like it might be
related to your problem. Try this patch:
http://archives.postgresql.org/pgsql-committers/2006-11/msg00121.php
regards, tom lane