To: Arjen van der Meijden <acmmailing@tweakers.net>
cc: Dave Dutcher <dave@tridecap.com>, nikolay@samokhvalov.com,
	pgsql-performance@postgresql.org
Subject: Re: Query plan for "heavy" SELECT with "lite" sub-SELECTs 
In-reply-to: <454B0E3C.4060600@tweakers.net> 
References: <e431ff4c0611020307x1fca3a14w16f60dbd80d0e856@mail.gmail.com>
	<012a01c6fe8a$bee330e0$8300a8c0@tridecap.com>
	<20061102191555.GL25444@alvh.no-ip.org>
	<454B0E3C.4060600@tweakers.net>
Comments: In-reply-to Arjen van der Meijden <acmmailing@tweakers.net>
	message dated "Fri, 03 Nov 2006 10:39:08 +0100"
Date: Fri, 03 Nov 2006 09:50:53 -0500
Message-ID: <1994.1162565453@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Arjen van der Meijden <acmmailing@tweakers.net> writes:
> ... Rewriting it to something like this made the last iteration about as 
> fast as the first:

> SELECT docid, (SELECT work to be done for each document)
> FROM documents
> WHERE docid IN (SELECT docid FROM documents
> 	ORDER BY docid
> 	LIMIT 1000
> 	OFFSET ?
> )

The reason for this, of course, is that the LIMIT/OFFSET filter is the
last step in a query plan --- it comes *after* computation of the SELECT
output list.  (So does ORDER BY, if an explicit sort step is needed.)
So if you have an expensive-to-compute output list, a trick like Arjen's
will help.  I don't think you can use an "IN" though, at least not if
you want to preserve the sort ordering in the final result.

			regards, tom lane