On Wed, Jan 5, 2011 at 6:08 PM, u235sentinel <u235sentinel@gmail.com> wrote:
> I'm tracking a problem with our tables being bloated and was curious if
> someone regularly kills autovacuum jobs, will autovacuum later reattempt to
> vacuum the table it was killed under?
>
> I've made autovacuum more aggressive and given more worker threads. Yet for
> some reason we're not keeping up.
Can you show us what you've changed to make autovac more aggressive?
You might want to make sure you've lowered
autovacuum_vacuum_cost_delay (mine's at 0 as we have 28 or so disks in
a single RAID-10 and lots of spare IO). Also raise
autovacuum_vacuum_cost_limit much higher (mine's at 5000). Assuming
these tables get updated a lot, I'm gonna guess that you're not having
a problem with too high of threshold settings but with auto-vac
keeping up. But it's just a guess.
Also, what do you get from things like iostat, vmstat, and sar as
regards your IO utilization and such? If your IO system is always at
100% then more aggressive vacuuming isn't gonna do much, because
you'll always be behind.