On Mon, Nov 01, 2010 at 08:18:24PM +0100, Cédric Villemain wrote:
> 2010/11/1 hubert depesz lubaczewski <depesz@depesz.com>:
> > On Mon, Nov 01, 2010 at 07:18:22PM +0100, Filip Rembiałkowski wrote:
> >> 2010/11/1 hubert depesz lubaczewski <depesz@depesz.com>:
> >>
> >> > as I understand, max number of xlog files in pg_xlog should be ( 1 + 2 *
> >> > checkpoint_segments ).
> >>
> >> why?
> >>
> >> for a server overloaded with R/W transactions, it's possible to go beyond this.
> >> checkpoints just do not keep up.
> >> right now I have an 8.3 with checkpoint_segments=3, constantly running
> >> pgbench and I see 8 WAL segments.
> >
> > you will notice in the logs that the system doesn't look like very
> > loaded.
> > i mean - there is fair amount of work, but nothing even resembling
> > "overloaded".
>
> There exists some checkpoint which occur more frequently than perhaps
> expected. (less than 15 minutes)...
>
> The logline about checkpoint might be usefull. Still I wonder what
> your question is exactly ?
why the number of wal segments is larger than 2n + 1
Best regards,
depesz
--
Linkedin: http://www.linkedin.com/in/depesz / blog: http://www.depesz.com/
jid/gtalk: depesz@depesz.com / aim:depeszhdl / skype:depesz_hdl / gg:6749007
