On ven, gen 11, 2002 at 03:16:12 -0500, Tom Lane wrote:
> Yes, I would like a copy of your script.
i will package script and 'instructions' soon.
i've got also two snapshots of db after the crash (tar files of postgres
directory)
>
> I just found and fixed a problem that could lead to this kind of failure
> (reference to an already-deleted CLOG entry). However, I'm not sure
yesterday i spent the afternoon doing a lot of test... now, i think you
don't need the details because...
> that the problem I fixed could affect VACUUM itself. Please update from
you was right, and there's no need of vacuum db to crash the postmaster...
doing a fresh installation, a fresh initdb, starting my script (and a
second bash script that does nothing more than a simple insert) after about
1 milion transactions i can see the problem (after changed sources as you told
me), without vacuumdb.
> cvs (make sure you get src/backend/utils/time/tqual.c v 1.46) and then
> see if you still see the problem. If you do, a backtrace from the point
> where elog() is called would be really helpful.
tomorrow i will redo all my tests, but i guess now the bug is fixed.
thanks a lot for your work,
andrea
n.b.: when you'll come in italy i'll be happy to be your host...
