Skip site navigation (1) Skip section navigation (2)

Re: SELECT LIMIT 1 VIEW Performance Issue

From: K C Lau <kclau60(at)netvigator(dot)com>
To: Simon Riggs <simon(at)2ndquadrant(dot)com>
Cc: pgsql-performance(at)postgresql(dot)org
Subject: Re: SELECT LIMIT 1 VIEW Performance Issue
Date: 2005-09-22 10:40:46
Message-ID: 6.2.1.2.0.20050922181009.02c36ca8@localhost (view raw or flat)
Thread:
Lists: pgsql-performance
We use similar views as base views throughout our OLTP system to get the 
latest time-based record(s). So it is quite impossible to use summary 
tables etc. Are there other ways to do it?

The subquery would pinpoint the record(s) with the composite primary key. 
Both MS Sql and Oracle do not have such performance problem. So this 
problem is effectively stopping us from migrating to PostgreSQL.

Any suggestions would be most appreciated.

Best regards,
KC.

At 16:40 05/09/22, Simon Riggs wrote:
>On Thu, 2005-09-22 at 12:21 +0800, K C Lau wrote:
>
> > Investigating further on this problem I brought up in June, the following
> > query with pg 8.0.3 on Windows scans all 1743 data records for a player:
> >
> > esdt=> explain analyze select PlayerID,AtDate from Player a
> >   where PlayerID='22220' and AtDate = (select b.AtDate from Player b
> >   where b.PlayerID = a.PlayerID order by b.PlayerID desc, b.AtDate desc
> > LIMIT 1);
> >
>
> >   Total runtime: 51.133 ms
> >
> > Using a static value in the subquery produces the desired result below, 
> but
> > since we use views for our queries (see last part of this email), we 
> cannot
> > push the static value into the subquery:
> >
> > esdt=> explain analyze select PlayerID,AtDate from Player a
> >   where PlayerID='22220' and AtDate = (select b.AtDate from Player b
> >   where b.PlayerID = '22220' order by b.PlayerID desc, b.AtDate desc 
> LIMIT 1);
>
> >   Total runtime: 0.149 ms
> >
> > The Player table has a primary key on PlayerID, AtDate. Is there a way to
> > stop the inner-most index scan looping all 1743 data records for that
> > player?  Is that a bug or known issue?
>
>Currently the planner can't tell whether a subquery is correlated or not
>until it has planned the query. So it is unable to push down the
>qualification automatically in the way you have achieved manually. The
>new min() optimisation doesn't yet work with GROUP BY which is what you
>would use to reformulate the query that way, so no luck that way either.
>
>If you don't want to do this in a view, calculate the values for all
>players at once and store the values in a summary table for when you
>need them.
>
>Best Regards, Simon Riggs
>
>
>
>---------------------------(end of broadcast)---------------------------
>TIP 6: explain analyze is your friend


In response to

Responses

pgsql-performance by date

Next:From: Simon RiggsDate: 2005-09-22 12:48:04
Subject: Re: SELECT LIMIT 1 VIEW Performance Issue
Previous:From: Simon RiggsDate: 2005-09-22 08:40:56
Subject: Re: SELECT LIMIT 1 VIEW Performance Issue

Privacy Policy | About PostgreSQL
Copyright © 1996-2014 The PostgreSQL Global Development Group