On Wed, Jan 27, 2016 at 11:54:37PM +0100, Ivan Voras wrote:
> So, question #1: WTF? How could this happen, on a regularly vacuumed
> system? Shouldn't the space be reused, at least after a VACUUM? The issue
> here is not the absolute existence of the bloat space, it's that it's
> constantly growing for *system* tables.
Some system tables (particularly pg_attribute) are heavily changed by
a lot of temp table use. You need to amp up the vacuum frequency on
them, and have a lot of workers, or you don't get to them until it's
too late.
> Question #2: What can be done about it?
You may end up taking an outage in effect, because you need to compact
them at least once. If you can flip to a replica, that is the easiest
way to fix it.
A
--
Andrew Sullivan
ajs@crankycanuck.ca