To: pgsql-performance@postgresql.org
Subject: Re: [GENERAL] Query is not using index when it should
References: <80oeh2tm7x.fsf@junk.nocrew.org>
	<20041210182453.G92467@megazone.bigpanda.com>
	<80d5xgudg6.fsf@junk.nocrew.org> <20041211143213.GA2659@uio.no>
From: tomas@nocrew.org (Tomas =?iso-8859-1?q?Sk=E4re?=)
Date: 12 Dec 2004 09:32:25 +0100
In-Reply-To: <20041211143213.GA2659@uio.no>
Message-ID: <806537udba.fsf@junk.nocrew.org>
Lines: 38
User-Agent: Gnus/5.0808 (Gnus v5.8.8) Emacs/20.7
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit

"Steinar H. Gunderson" <sgunderson@bigfoot.com> writes:

> On Sat, Dec 11, 2004 at 03:17:13PM +0100, Tomas Skäre wrote:
> > select c.* from cjm_object c
> >  inner join
> >   (select max(timestamp) as timestamp,objectid,field from cjm_object
> >    group by objectid,field) t
> >   using(timestamp,objectid,field)
> >  where 1=1 and data is not null
> >  order by objectid,field;
> 
> Usually, SELECT max(field) FROM table is better written in PostgreSQL as
> SELECT field FROM table ORDER field DESC LIMIT 1.

Well, my subquery doesn't return just one row, but one for each
objectid,field combination in the table. I could rewrite it to
something like this:

select c.* from cjm_object c
where exists (select timestamp from cjm_object t
              where c.objectid=t.objectid
              and c.field=t.field
              order by timestamp desc limit 1)
and data is not null
order by objectid;

But that seems to be even slower, even if it can use an index scan in
the subquery. Also it doesn't give the same result set, but I haven't
looked into what's wrong yet.

> I don't see the point of "where 1=1", though...

It's just because the actual query is generated by a program, and it's
easier to always have "where 1=1" and then add optional conditions
with "and ...".


Tomas