Re: pg_upgrade and statistics

On Tue, Mar 13, 2012 at 05:33:29PM -0500, Kevin Grittner wrote:
> Bruce Momjian <bruce@momjian.us> wrote:
>  
> > What is the target=10 duration?  I think 10 is as low as we can
> > acceptably recommend.  Should we recommend they run vacuumdb
> > twice, once with default_statistics_target = 4, and another with
> > the default?
>  
> Here are the results at various settings.
>  
> 1   :  172198.892 ms
> 2   :  295536.814 ms
> 4   :  474319.826 ms
> 10  :  750458.312 ms
> 100 :  3433794.609 ms

Thanks, good numbers to know.

> I'm not sure what's best for a general approach to the problem.  For
> my own part, I'd be inclined to cherry-pick tables if I were in a
> hurry.
>  
> I hope we at least bring over relpages and reltuples, to give the
> optimizer *some* clue what it's looking at.  I wouldn't thing those
> would be changing semantics or format very often.

True, but we don't migrate them either.

This is the exact same problem you would have restoring a pg_dump
backup.  The improvement needs to go into pg_dump, and then pg_upgrade
can make use of it.

Another idea is to just copy over pg_statistic like we copy of
pg_largeobject now, and force autovacuum to run.

--  Bruce Momjian  <bruce@momjian.us>        http://momjian.us EnterpriseDB
http://enterprisedb.com
 + It's impossible for everything to be true. +