On Mon, Oct 19, 2009 at 11:27 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Scott Marlowe <scott.marlowe@gmail.com> writes:
>> All of this is completely besides the point that a database that's
>> been shutdown immediately / had the power cord yanked comes back up
>> and doesn't start autovacuuming automatically, which seems a
>> non-optimal behaviour.
>
> It'll start as soon as you've modified enough rows. The absolute worst
> case behavior is that table bloat reaches twice the level it would have
> otherwise, or pg_statistic data becomes twice as out of date as it would
> have otherwise.
That could be a pretty bad worst case scenario for certain types of
tables / usage patterns.
How bad can the affect of out of date pg_statistic data be? Is it
likely to turn a hash agg into a nested loop and take a query from
seconds to minutes? If so, then that's pretty bad.
> Now, if your server's MTBF is less than the autovac interval, you could
> indeed have an accumulating problem ... but I suggest that in that
> situation you've got other issues to fix.
True. Still very much beside the point.
