> I wonder whether you saw some variant of the deadlock reported in bug
> #3883 --- see discussion here:
> http://archives.postgresql.org/pgsql-hackers/2008-01/msg00873.php
> The only known way of getting into that problem is fixed in 8.3,
> but a potential for silent deadlock is definitely still there.
> I don't however have any idea how such a deadlock could occur without
> at least some processes showing as 'waiting' in pg_stat_activity.
>
> Do you have any idea what the transactions that were blocked at
> COMMIT had been doing?
Should have just been selects and updates, and perhaps some inserts
and deletes - nothing to out of the ordinary or complex.
If I can get this to happen consistently, I'll set up a logger to
capture all of the SQL statements leading up to the freeze.
>
> Are there any foreign key constraints leading *to* this table?
Nope.
Unfortunately, I had been running for a couple of weeks steady before
I got the error - so reproducing it may not be quick. But I'm going
to set up a couple of parallel tests on more systems, and see if I can
happen into it again sooner.
Oh, and I don't think though should have anything to do with it, but I
did discover that fsync was turned off on this postgres instance.
Someone else didn't clean up after a different test, and I forgot to
change it back.
Thanks,
Dan