Re: query planning different in plpgsql?

On Mon, 26 Oct 2009 14:09:49 -0400, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> "Michal J. Kubski" <michal(dot)kubski(at)cdt(dot)pl> writes:
>> [ function that creates a bunch of temporary tables and immediately
>> joins them ]
>
> It'd probably be a good idea to insert an ANALYZE on the temp tables
> after you fill them. The way you've got this set up, there is no chance
> of auto-analyze correcting that oversight for you, so the planner will
> be planning the join "blind" without any stats. Good results would only
> come by pure luck.
>
> regards, tom lane

Hi,

Apologies for late response. Thanks a lot: ANALYZE seem to help it! I
still sometimes
get long query runs though. As far as I understand using index over
sequential scan
on joins should be faster. Could it be possible that the query planner
decides
to use seqscan instead of index scan on some random occasions?

Thanks,
Michal

--
I hear and I forget. I see and I believe. I do and I understand.
(Confucius)