Message-ID: <4635F2E1.3090602@archonet.com>
Date: Mon, 30 Apr 2007 14:45:05 +0100
From: Richard Huxton <dev@archonet.com>
User-Agent: Thunderbird 2.0.0.0 (X11/20070326)
MIME-Version: 1.0
To: Andreas Haumer <andreas@xss.co.at>
Cc: Guillaume Cottenceau <gc@mnc.ch>,
	pgsql-performance@postgresql.org
Subject: Re: Query performance problems with partitioned tables
References: <4635D678.3050205@xss.co.at> <87ps5mgghp.fsf@meuh.mnc.lan>
	<4635EF3A.1070204@xss.co.at>
In-Reply-To: <4635EF3A.1070204@xss.co.at>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Andreas Haumer wrote:
> 
> I think the planner could do the following:
> 
> a) It could make a better decision in which direction to scan
>    the partitions (depending on sort order involved in the query)
> 
> b) It could stop scanning as soon as there can not be any further
>    resulting row according to the CHECK constraints given on the tables.
[snip]
> Perhaps the logic to implement this is complex, but IMHO
> it _should_ be doable (and proofable), shouldn't it?

Ah, it might be do-able for some subset of cases, but is it 
cost-effective to check for in *all* cases? Don't forget the constraints 
and where clauses can be arbitrarily complex.
-- 
   Richard Huxton
   Archonet Ltd