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

Re: performance with query

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: "Alberto Dalmaso" <dalmaso(at)clesius(dot)it>, "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: <pgsql-performance(at)postgresql(dot)org>
Subject: Re: performance with query
Date: 2009-06-18 19:54:38
Message-ID: 4A3A552E0200002500027DB3@gw.wicourts.gov (view raw or flat)
Thread:
Lists: pgsql-performance
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote: 
> "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
>> With all the optimizer options on, and the from_collapse_limit and
>> join_collapse_limit values both set to 100, run an EXPLAIN (no
>> ANALYZE) on your big problem query.  Let us know how long the
>> EXPLAIN runs.  If it gets any errors, copy and paste all available
>> information.  (General descriptions aren't likely to get us very
>> far.)  Since EXPLAIN without ANALYZE only *plans* the query, but
>> doesn't run it, it should not take long to do this.
> 
> One issue here is that with the collapse limits cranked up to more
> than geqo_threshold, he's going to be coping with GEQO's partially-
> random plan selection; so whatever he reports might or might not be
> especially reflective of day-to-day results.  I'm tempted to ask
> that he also push up geqo_threshold.
 
In an earlier post[1] he said that he had geqo turned off.  It does
pay to be explicit, though; I'd hate to assume it's of if he's been
changing things.
 
Alberto, please ensure that you still have geqo off when you run the
test I suggested.  Also, I see that I didn't explicitly say that you
should send the ANALYZE output, but that's what would be helpful.
 
> It's possible that that *will* send the planning time to the moon;
> but it would certainly be worth trying, to find out what plan is
> produced.
 
Agreed.  What plan is produced, and how long that takes.  (And whether
he gets an out of memory error.)  I figured it was best to get a clear
answer to those before moving on....
 
-Kevin
 
[1]
http://archives.postgresql.org/pgsql-performance/2009-06/msg00186.php

In response to

pgsql-performance by date

Next:From: Kenneth MarshallDate: 2009-06-18 20:01:02
Subject: Re: Strange performance response for high load times
Previous:From: Peter AlbanDate: 2009-06-18 19:42:47
Subject: Re: Strange performance response for high load times

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