(re-adding hackers)
On Fri, Feb 1, 2013 at 2:46 PM, Vlad Bailescu <vlad@mojitosoftware.com> wrote:
>
> I'm pretty sure the io is from the autovacuum on master table because it's
> last_autovacuum stats update almost every minute
That only proves that the master table is being vacuumed, but does not
necessarily say that others are not. I would suggest turning auto
vacuum logging on.
> and iotop shows something
> like:
>
> Total DISK READ: 5.80 M/s | Total DISK WRITE: 115.85 K/s
> TID PRIO USER DISK READ DISK WRITE SWAPIN IO> COMMAND
> 7681 be/4 postgres 5.93 M/s 0.00 B/s 0.00 % 63.62 % postgres:
> autovacuum worker process fleet
Whats interesting is that there is a lot of reads happening, but very
little writes. I can see a scenario when this can happen even with
visibility maps because vacuum honors visibility map only if it can
skip 32 blocks at stretch, but your other claim that this goes on for
90 seconds even though the table is just couple of MBs is hard to
justify. How does the auto vacuum settings look like ? I wonder if you
have scaled down naptime and threshold just too low.
Thanks,
Pavan
--
Pavan Deolasee
http://www.linkedin.com/in/pavandeolasee