To: Jeff Davis <pgsql@j-davis.com>
cc: "Jim C. Nasby" <jim@nasby.net>, Dimitri Fontaine <dim@dalibo.com>,
	pgsql-performance@postgresql.org, Ioana Danes <ioanasoftware@yahoo.ca>
Subject: Re: Postgresql 8.1.4 - performance issues for select on 
In-reply-to: <1161210735.31645.277.camel@dogma.v10.wvs> 
References: <20061018210201.56435.qmail@web55907.mail.re3.yahoo.com>
	<200610182319.41920.dim@dalibo.com>
	<1161207230.31645.266.camel@dogma.v10.wvs>
	<20061018221046.GG85041@nasby.net>
	<1161210735.31645.277.camel@dogma.v10.wvs>
Comments: In-reply-to Jeff Davis <pgsql@j-davis.com>
	message dated "Wed, 18 Oct 2006 15:32:15 -0700"
Date: Wed, 18 Oct 2006 19:05:17 -0400
Message-ID: <2694.1161212717@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Jeff Davis <pgsql@j-davis.com> writes:
> If PostgreSQL could sort the result of a union by merging the results of
> two index scans, I think the problem would be solved. Is there something
> preventing this, or is it just something that needs to be added to the
> planner?

It's something on the wish-list.  Personally I'd be inclined to try to
rewrite the query as a plain MAX() across rewritten per-table indexed
queries, rather than worry about mergesort or anything like that.
There wasn't any very good way to incorporate that idea when planagg.c
was first written, but now that the planner has an explicit notion of
"append relations" it might be relatively straightforward.

			regards, tom lane