Alvaro Herrera <alvherre@commandprompt.com> writes:
> Casey Duncan wrote:
>> I'm curious how template0 got stomped on.
> Heh :-) Sorry, they are all my bugs. I guess you should be throwing
> stones at me or something.
The pre-8.1 theory was that template0 is (supposed to be) cleanly frozen
and hence never needs vacuumed at all. The post-8.1 theory is that
template0 gets autovacuumed when necessary to prevent wraparound, just
like every other database. 8.1 unfortunately is somewhere in the
middle, because under circumstances-I-don't-remember-at-the-moment,
autovacuum might decide to process template0 and then leave non-frozen
XIDs therein. Which is a problem because the clog-truncation logic
didn't think it needed to consider template0 when deciding if old clog
segments could be thrown away. We live and learn.
The real bottom line here, and one I'll reiterate every chance I get,
is that we don't make updates to back branches because we're too bored
to have anything else to do. If you're on 8.1.5, and the current
release in that branch is 8.1.8, then you're missing some bug fixes
that are probably significant.
regards, tom lane