Date: Sat, 5 May 2007 20:40:53 -0400
From: Andrew Sullivan <ajs@crankycanuck.ca>
To: pgsql-sql@postgresql.org
Subject: Re: Sequence vs. Index Scan
Message-ID: <20070506004053.GA22890@phlogiston.dyndns.org>
References: <bf05e51c0705051405p29ee139fn30c9979f3772a1d4@mail.gmail.com>
	<20070505213009.GA22573@phlogiston.dyndns.org>
	<bf05e51c0705051500p3e3de711na0d76ae1ad463775@mail.gmail.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <bf05e51c0705051500p3e3de711na0d76ae1ad463775@mail.gmail.com>
User-Agent: Mutt/1.5.9i

On Sat, May 05, 2007 at 05:00:53PM -0500, Aaron Bono wrote:
> 
> They have different data.  The fast one has about 150 rows and the  slow one
> has about 40 rows.  The field in question here, the branch_id, is a
> BIGSERIAL in both.

I'd be astonished if a table of 40 rows ever got index scanned.  It's
probably more efficient to read the whole table.  But it seems your
case may be strange.  

> We don't allow deletes and updates are fairly infrequent.  I also did a
> vacuum analyze to no effect.

How do you "not allow" deletes?  Does this happen because of a failed
foreign key?  If so, you can end up with dead tuples.  I'd look at
the output of VACUUM VERBOSE to make sure you don't have a lot of
dead tuples.  That said, I wonder if fiddling with the statistics on
your tables might help.

A

-- 
Andrew Sullivan  | ajs@crankycanuck.ca
In the future this spectacle of the middle classes shocking the avant-
garde will probably become the textbook definition of Postmodernism. 
                --Brad Holland