To: "Marc G. Fournier" <scrappy@postgresql.org>
cc: pgsql-sql@postgresql.org
Subject: Re: timestamp query doesn't use an index ... 
In-reply-to: <20060520223704.T1114@ganymede.hub.org> 
References: <20060520223704.T1114@ganymede.hub.org>
Comments: In-reply-to "Marc G. Fournier" <scrappy@postgresql.org>
	message dated "Sat, 20 May 2006 22:42:00 -0300"
Date: Sun, 21 May 2006 11:31:02 -0400
Message-ID: <5287.1148225462@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Marc G. Fournier" <scrappy@postgresql.org> writes:
> I'm trying to figure out some way to speed up the following query:

>    select ps2.page_id, ps2.template_component_id, max(ps2.start_time)
>      from page_schedule ps2
>     where ps2.start_time at time zone 'MST7MDT' <= '2006-5-17 8:9:18'
> group by ps2.page_id, ps2.template_component_id

> Is there some other way I can either write above query *or* do an index, 
> such that it will use the index?

One-sided inequalities frequently *shouldn't* use an index, because
they're retrieving too much of the table.  Are you sure this is fetching
only a small fraction of the table?  Are you using PG 8.1 (8.1 would be
likely to try to use a bitmap indexscan for this)?

You could experiment with enable_seqscan = off to see if the planner is
actually wrong about its choice.  If so, reducing random_page_cost might
be the best permanent solution.

			regards, tom lane