Message-ID: <772C0DC9C4A04AA4939EB50F2707A791@andrusnotebook>
From: "Andrus" <kobruleht2@hot.ee>
To: "Scott Carey" <scott@richrelevance.com>,
	<pgsql-performance@postgresql.org>, "PFC" <lists@peufeu.com>
References: <gh38cg$1544$1@news.hub.org> <op.uljfbad8cigqcu@soyouz>
	<B52A6C218D2145378337CFDC89699D99@andrusnotebook>
	<op.uljghcfxcigqcu@soyouz>
	<95776070DDC841978DF8335DCF59D6BB@andrusnotebook>
	<op.uljja8egcigqcu@soyouz>,
	<72246E216337406EAD285F5A45533672@andrusnotebook>
	<BDFBB77C9E07BE4A984DAAE981D19F961ACA17D9EF@EXVMBX018-1.exch018.msoutlookonline.net>
In-Reply-To: 
 <BDFBB77C9E07BE4A984DAAE981D19F961ACA17D9EF@EXVMBX018-1.exch018.msoutlookonline.net>
Subject: Re: analyzing intermediate query
Date: Tue, 2 Dec 2008 20:58:43 +0200
MIME-Version: 1.0
Content-Type: text/plain; format=flowed; charset="iso-8859-1";
	reply-type=original
Content-Transfer-Encoding: 7bit

Scott,

>Often times, switching an inner subselect that requires a distinct to a 
>group by on that column yields better results.  In this case, the IN should 
>be equivalent, so it probably will not help.  This would look like:

SELECT dok.*
 FROM dok
JOIN  (SELECT dokumnr FROM  temptbl GROUP BY dokumnr ) x USING(dokumnr);

Thank you. This may be great idea.
I changed my query to use GROUP BY instead of DISTINCT

>Whether that hepls depends on how big dokumnr is and where the query 
>bottleneck is.

I'm wondering how this can solve the issue when there is single or few 
dokumnr columns.
Planner still thinks that temptbl projection contains 1000 rows and uses seq 
scan instead of using bitmap index on dok table.

I tried

SELECT dok.*
 FROM dok
JOIN  (SELECT dokumnr FROM  temptbl GROUP BY dokumnr ANALYZE ) x 
USING(dokumnr);

but got error.

> Note there are subtle differences between DISTINCT and GROUP BY with 
> respect to nulls.

dokumnr is int type and is not null always.

Andrus.