On Oct 2, 2006, at 17:50 , Tom Lane wrote:
> Alexander Staubo <alex@purefiction.net> writes:
>> I have a production PostgreSQL instance (8.1 on Linux 2.6.15) that
>> seems to be writing data to disk at rates that I think are
>> disproportional to the update load imposed on the database. I am
>> looking for ways to determine the cause of this I/O.
>
> Are you sure that iostat is to be trusted?
No. :) But iostat reads directly from /dev/diskstats, which should be
reliable. Of course, it still doesn't say anything about which
process is doing the writing; for that I would need to install the
atop kernel patches or similar.
...
> The read numbers in
> particular look suspiciously uniform ... it would be a strange
> query load that would create a read demand changing less than 1%
> from hour to hour, unless perhaps that represented the disk's
> saturation point, which is not the case if you're not seeing
> obvious performance problems.
They are not uniform at all -- they correlate perfectly with the web
traffic; it just so happens that the samples I quoted were from peak
hours. Take a look at the Munin graph. (The spikes correspond to
scheduled maintenance tasks like backups.)
Alexander.