"Markus Wollny" <Markus.Wollny@computec.de> writes:
> Sorry for the quick updates to my own messages, but I didn't want to
> lean back and wait - so I took to more aggressive measures. All my
> other databases in this cluster are fine - and the 'postgres' database
> doesn't seem to do anything really useful except being the default
> database. I dropped it and recreated it with template1 as template,
> afterwards I could start up my cluster with no problems
> whatsoever.
Yeah, if there were no other problems apparent in pg_database I was
going to suggest that as a recovery method.
> I'd still like to find out what exactly happened here so I
> can prevent the same from happening again in the future.
Me too. It would seem that something did a vacuum of postgres
with a strange choice of xid cutoff, but I can't think of what
would cause that.
Do you ever do VACUUM FREEZE on your databases?
regards, tom lane