On 2013-08-19 15:25:38 -0400, Tom Lane wrote:
> Andres Freund <andres@2ndquadrant.com> writes:
> > Hm. Is a check like that actually sufficient? The idea of setting
> > stats_temp_directory to /dev/shm/postgres or similar in all of several
> > clusters on one machine doesn't seem to be that far fetched.
>
> Hm. We could fairly easily have the lockfile management code also
> write a postmaster.pid file into the stats_temp_directory, thus claiming
> it in the same way as we do the $PGDATA dir itself. Not sure it's worth
> the trouble though, since we've not heard any field reports of this sort
> of thing.
The reason I think it's more likely is that pg_stats_directory, to be
useful, really needs something like a ramdisk/tmpfs. Which is annoying
to create in a persistent fashion...
Very likely doing so would cause hard to diagnose planner issues using
completely absurd statistics. Not sure how often it would get properly
diagnosed.
But:
> Maybe we're overreacting to this issue for stats_temp_directory,
> and tightening up the deletion code is a sufficient fix.
You very well might be right. Just wanted to raise the issue.
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services