Re: [Sd-log] Re: Endless loop in ExecNestLoop

Hi,

On Monday 30 January 2006 17:42, Tom Lane wrote:
> > Version: 8.0.2
> I don't think so ... neither bitmap scans nor slot_getattr exist in 8.0.

oops. it is 8.1.2 ..

> Not a lot. You need to find out what the difference is between the
> cases where the query runs quickly and those where it doesn't. I'm
> betting that the planner is choosing a very bad plan in the slow cases,
> but there's not a lot of evidence here to show what or why.

it is exactly the same query. I rather think that there is some kind of memory
corruption causing that error, so that the query runs fine in a "fresh"
postgres process but fails when e.g. malloc() returns a block which has
already old data in it or the stack contains already data from an old stack
fragment. We hunted a bug with simmilar symptomps a while ago:

http://archives.postgresql.org/pgsql-bugs/2003-08/msg00051.php

> Do you have geqo_threshold set to less than its default value?

We have disabled geqo on that system.

> sometimes execute the query with different sets of parameters?
> Either of these might lead to changes of plan.

no. it is exactly the same query that sometimes executes fast and sometimes
hangs for hours until we kill the process.

yours,
- clifford

--
: Clifford Wolf                      Tel +43-1-8178292-00 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Schönbrunnerstr 244, 1120 Vienna, Austria    http://www.linbit.com :