Re: Recommendations for set statistics - Mailing list pgsql-performance

From Josh Berkus
Subject Re: Recommendations for set statistics
Date
Msg-id 200505130922.11582.josh@agliodbs.com
Whole thread Raw
In response to Re: Recommendations for set statistics  (Christopher Browne <cbbrowne@acm.org>)
List pgsql-performance
Chris,

> It is widely believed that a somewhat larger default than 10 would be
> a "good thing," as it seems to be fairly common for 10 to be too small
> to allow statistics to be stable.  But nobody has done any formal
> evaluation as to whether it would make sense to jump from 10 to:
>
>  - 15?
>  - 20?
>  - 50?
>  - 100?
>  - More than that?

My anecdotal experience is that if more than 10 is required, you generally
need to jump to at least 100, and more often 250.   On the other end, I've
generally not found any difference between 400 and 1000 when it comes to
"bad" queries.

I have an unfinished patch in the works which goes through and increases the
stats_target for all *indexed* columns to 100 or so.   However, I've needed
to work up a test case to prove the utility of it.

--
Josh Berkus
Aglio Database Solutions
San Francisco

pgsql-performance by date:

Previous
From: "Steinar H. Gunderson"
Date:
Subject: Re: PostgreSQL strugling during high load
Next
From: Tom Lane
Date:
Subject: Re: PostgreSQL strugling during high load