On Wed, Nov 7, 2012 at 1:58 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote:
> On Wed, Nov 7, 2012 at 1:42 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Lists <lists@benjamindsmith.com> writes:
>>
>>> ... because it
>>> occasionally causes transactions and queries to hang when an update
>>> causes a vacuum mid-day, effectively taking us offline randomly.
>>
>> I suspect this claim is based on ancient and no longer very relevant
>> experience.
>
> My experience is that if autovac is causing problems with stalled
> queries etc you're either A: running ancient pg versions (pre 8.3), B:
> Running WAY too aggressive settings in autovac (100 threads, no nap
> time, cost limit of 100000 etc.) or C: Your IO subsystem is absolute
> crap.
>
> On any modern server, default autovac settings from 8.3 and on should
> only have the possible problem of not being tuned aggressively enough.
Oh another failure scenario up there is that you're running DDL in
production, which is stalling behind an autovac, and in turn the two
are stalling other queries. This has happened for me once or twice on
more modern versions (8.3 and 8.4)