Date: Sat, 29 Oct 2005 10:05:27 -0600
From: Michael Fuhr <mike@fuhr.org>
To: Bruno Wolff III <bruno@wolff.to>, Martin Lesser <ml-pgsql@bettercom.de>,
	pgsql-performance@postgresql.org
Subject: Re: Effects of cascading references in foreign keys
Message-ID: <20051029160527.GA80135@winnie.fuhr.org>
References: <87acgszihk.fsf@nb-aspire.bettercom.de>
	<20051029142432.GA79557@winnie.fuhr.org>
	<20051029144947.GB21068@wolff.to>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20051029144947.GB21068@wolff.to>

On Sat, Oct 29, 2005 at 09:49:47AM -0500, Bruno Wolff III wrote:
> On Sat, Oct 29, 2005 at 08:24:32 -0600, Michael Fuhr <mike@fuhr.org> wrote:
> > My tests suggest that a lookup on the referring key is done only
> > if the referenced key is changed.  Here's an example from 8.1beta4;
> > I used this version because EXPLAIN ANALYZE shows triggers and the
> > time spent in them, but I see similar performance characteristics
> > in earlier versions.  I've intentionally not put an index on the
> > referring column to make lookups on it slow.
> 
> It looks like this feature was added last May, so I think it only applies
> to 8.1.

Earlier versions appear to have at least some kind of optimization.
Here's a test in 7.3.11 using the same tables I used in 8.1beta4,
although on a slower box.

test=> UPDATE foo SET x = 1 WHERE id = 100000;
UPDATE 1
Time: 32.18 ms

test=> UPDATE foo SET x = 1, id = 200000 WHERE id = 100000;
UPDATE 1
Time: 4144.95 ms

test=> DROP TABLE bar;
DROP TABLE
Time: 240.87 ms

test=> UPDATE foo SET x = 1, id = 100000 WHERE id = 200000;
UPDATE 1
Time: 63.52 ms

-- 
Michael Fuhr