To: arnaulist@andromeiberica.com
cc: pgsql-performance@postgresql.org
Subject: Re: Is it possible to speed this query up? 
In-reply-to: <44C79FED.5080702@andromeiberica.com> 
References: <44C79FED.5080702@andromeiberica.com>
Comments: In-reply-to Arnau <arnaulist@andromeiberica.com>
	message dated "Wed, 26 Jul 2006 19:01:33 +0200"
Date: Wed, 26 Jul 2006 16:24:47 -0400
Message-ID: <17741.1153945487@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Arnau <arnaulist@andromeiberica.com> writes:
> the explain analyze shouts the following:

The expensive part appears to be this indexscan:

>                       ->  Index Scan using pk_agndusrgrp_usergroup on 
> agenda_users_groups ug  (cost=0.00..123740.26 rows=2936058 width=30) 
> (actual time=0.101..61921.260 rows=2836638 loops=1)

Since there's no index condition, the planner is evidently using this
scan just to obtain sort order.  I think ordinarily it would use a
seqscan and then sort the final result, which'd be a lot faster if the
whole result were being selected.  But you have a LIMIT and it's
mistakenly guessing that only a small part of the table will need to be
scanned before the LIMIT is satisfied.

Bottom line: try dropping the LIMIT.  If you really need the limit to be
enforced on the SQL side, you could try declaring the query as a cursor
and only fetching 150 rows from it.

			regards, tom lane