Message-ID: <460CE775.8080908@archonet.com>
Date: Fri, 30 Mar 2007 11:33:25 +0100
From: Richard Huxton <dev@archonet.com>
User-Agent: IceDove 1.5.0.9 (X11/20061220)
MIME-Version: 1.0
To: Gaetano Mendola <mendola@bigfoot.com>
Cc: "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Subject: Re: Wrong plan sequential scan instead of an index one
References: <460CD941.6080303@bigfoot.com> <460CDCDC.7020901@archonet.com>
	<460CE359.8060208@bigfoot.com>
In-Reply-To: <460CE359.8060208@bigfoot.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Gaetano Mendola wrote:
> 
> Richard Huxton wrote:
>>
>> Now, why 19 rows from the subquery should produce such a large estimate
>> in the outer query I'm not sure. Any strange distribution of values on
>> pvcp?
> 
> I don't know what do you mean for strange, this is the distribution:
> 
> test=# select count(*) from t_oa_2_00_card;
>  count
> - --------
>  877682
> (1 row)
> 
> test=# select count(*), pvcp from t_oa_2_00_card group by pvcp;
>  count | pvcp
> - -------+------
> (92 rows)
> 
> 
> I think that estimate is something like:  877682 / 92 * 19

So if you actually had 19 matches for '%pi%' it might be a sensible plan 
then. I'm afraid I don't know of any way to improve PG's prediction on 
how many matches you'll get for a substring pattern though.

-- 
   Richard Huxton
   Archonet Ltd