On Fri, Oct 22, 2010 at 11:17 AM, Kevin Grittner
<Kevin.Grittner@wicourts.gov> wrote:
> I've not seen anything remotely like that. I wonder what's
> different....
I did a test with the same postgres build (and autovacuum_naptime = 1
as well) on the same OS with only the system databases (postgres,
template0, template1), and after the weekend the memory utilization
was acceptable:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
14879 pgsql 15 0 312m 177m 1108 S 0.0 8.8 0:24.43 postgres
However on the other machine which has 136 total databases (133 user +
the same 3 system ones listed above), the kernel had to kill off the
postgres process since the box ran out of memory (the box has 8GB ram
+ 30GB swap), and I see it climbing again.
It seems like the problem is a function of how many databases there are.
> Is there anything unusual in the logs?
I set log_autovacuum_min_duration = 0 but that doesn't even seem to
show that anything is being vacuumed. The only thing unusual in the
logs was a result of the box running out of memory and the kernel
killing off processes.