Re: *_collapse_limit, geqo_threshold

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: *_collapse_limit, geqo_threshold
Date: 2009-07-09 15:37:41
Message-ID: 4A55C875020000250002864F@gw.wicourts.gov
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Resending to list due to failure:
451 4.3.5 Server configuration problem

Joshua Tolley <eggyknap(at)gmail(dot)com> wrote:

> Entire stored plans, or predetermined seeds for GEQO's random number
> generator all boil down to saying, "I want you to use this plan
> henceforth and forever."

A predetermined seed for geqo's random number generator only
guarantees that you get the same plan for the same query as long as
the statistics remain the same. After the next ANALYZE, you may still
get a different plan.

> Do we know that GEQO plans are, in reality, less stable than than
> usual planner?

Yeah, I had a complaint of a slow query. When I ran EXPLAIN ANALYZE
on it I ran it several times (as I usually do, to establish what
happens with and without caching). I got different plans. So I ran
the query a bunch of times and got a decent plan about 90% of the
time, and an awful one about 10% of the time. It was clearly a geqo
effect, so I didn't post on it. (Why post? The product was clearly
operating as intended.) There was an easy solution; I disabled geqo.
I'm sure it's useful to some; we haven't missed it.

-Kevin

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Andres Freund 2009-07-09 15:39:26 Re: *_collapse_limit, geqo_threshold
Previous Message Andres Freund 2009-07-09 15:00:42 Re: *_collapse_limit, geqo_threshold - example schema