On Thu, Jun 08, 2006 at 09:13:33AM -0700, Joshua D. Drake wrote:
> Tom Lane wrote:
> >"Joshua D. Drake" <jd@commandprompt.com> writes:
> >>I was recently involved in a project where we had to decrease the
> >>checkpoint_timeout . The problem was, that the database was performing
> >>so many transactions that if we waiting for 5 minutes, checkpoint would
> >>take entirely too long.
> >
> >Seems like the correct fix for that is to make the bgwriter more
> >aggressive. Narrowing the checkpoint spacing is a pretty horrid answer
> >because of the resulting increase in full-page-image WAL traffic.
>
> Well we did that as well. Here are the basic symptons:
>
> During normal processing which contained about 250 connections
> everything was fine. A checkpoint would start and connections would
> start piling up, sometimes breaking 1000.
>
> We narrowed that down to users having to wait longer for query execution
> so instead of just reusing connections new connections had to be
> initiated because the existing connections were busy.
>
> We tried many different parameters, and bgwriter did significantly help
> but the only "solution" was to make checkpoints happen at a much more
> aggressive time frame.
>
> Modify bgwriters settings and the checkpoint actually increased our
> velocity by about 70% by the time we were done. Bgwriter was definitely
> the largest chunk of that although other parameters combined outweighed
> it (effective_cache, shared_buffers etc...).
Did you try increasing the checkpoint interval, in the hopes that it
would allow the bgwritter enough extra time to get everything pushed
out?
--
Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com
Pervasive Software http://pervasive.com work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461