Date: Sat, 28 Jun 2008 21:19:31 +0200
From: "Steinar H. Gunderson" <sgunderson@bigfoot.com>
To: Moritz Onken <onken@houseofdesign.de>
Cc: pgsql-performance@postgresql.org
Subject: Re: Planner should use index on a LIKE 'foo%' query
Message-ID: <20080628191931.GB13013@uio.no>
Mail-Followup-To: Moritz Onken <onken@houseofdesign.de>,
	pgsql-performance@postgresql.org
References: <287E4428-BECC-46D6-BC47-EA94EA0121F0@houseofdesign.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <287E4428-BECC-46D6-BC47-EA94EA0121F0@houseofdesign.de>
User-Agent: Mutt/1.5.18 (2008-05-17)

On Sat, Jun 28, 2008 at 06:24:42PM +0200, Moritz Onken wrote:
>  SELECT distinct url from item where url like 'http://www.micro%' limit 
> 10;

Here, the planner knows the pattern beforehand, and can see that it's a
simple prefix.
> select *
>  from result
>  where exists
>    (select * from item where item.url LIKE result.url || '%' limit 1)  
> limit 10;

Here it cannot (what if result.url was '%foo%'?).

Try using something like (item.url >= result.url && item.url <= result.url ||
'z'), substituting an appropriately high character for 'z'.

> The only explaination is that I don't use a constant when comparing the 
> values. But actually it is a constant...

It's not a constant at planning time.

Also note that you'd usually want to use IN instead of a WHERE EXISTS.

/* Steinar */
-- 
Homepage: http://www.sesse.net/