Please reply to the list, not directly to me.
> I don't think is that simple. The VERY SAME statement runs twice - one
> finishes in about 20 secs the other doesn't finish in 24 hours.
Yep, OK, so it's not just a planning or scaling issue.
Have you checked pg_locks ?
SELECT * FROM pg_locks;
What does pg_stat_activity indicate about the query?
SELECT * FROM pg_stat_activity;
> The plan might change from one execution to the other - is there a way
> to get the executed plan at runtime?
I think there is in 8.4, but I haven't moved up to it and tested yet.
Not in previous versions.
--
Craig Ringer