[Apologies if you have seen this before. I just discovered that posting
to the group via NNTP, at Teranews.com, apparently isn't working since
my posts aren't showing up in the mailing list archives.]
We are trying to run down some performance differences between our
production system (where we r
Jeff Boes <[EMAIL PROTECTED]> writes:
> For large (>1 million rows) tables
> which have a pretty high turn-over (average life span of a row is 3
> days), should there be any query performance differences whether you
> VACUUM FULL or not?
How often do you VACUUM? Do you have enough FSM space to su
On Thu, 2003-06-12 at 12:42, Tom Lane wrote:
> Jeff Boes <[EMAIL PROTECTED]> writes:
> > For large (>1 million rows) tables
> > which have a pretty high turn-over (average life span of a row is 3
> > days), should there be any query performance differences whether you
> > VACUUM FULL or not?
>
> H
Jeff Boes <[EMAIL PROTECTED]> writes:
> On Thu, 2003-06-12 at 12:42, Tom Lane wrote:
>> How often do you VACUUM?
> We've gone from daily, to twice daily, to several times during the "peak
> updates" period, and back to twice daily.
I suspect that the real problem here is the old open transactions
On Thu, 2003-06-12 at 13:16, Tom Lane wrote:
> > We're also ANALYZE-ing the largest 12-18 tables on a cycle: every twenty
> > minutes, a daemon wakes up and ANALYZEs until they're all done or two
> > minutes has elapsed, whichever comes first.
>
> That sounds a tad excessive; are the statistics r
Jeff Boes <[EMAIL PROTECTED]> writes:
> On Thu, 2003-06-12 at 13:16, Tom Lane wrote:
>> That sounds a tad excessive; are the statistics really changing that
>> fast?
> Well, I have some convincing evidence on this.
Ah. Well, you might consider analyzing just that table (maybe even just
its times