"Andy Shellam (Mailing Lists)" <andy.shellam-lists@mailnetwork.co.uk> writes:
> Is it enough to simply have re-copied in the base/xxx directory from the
> base backup, after the PITR recovery had completed (obviously any
> changes made to that database since the base backup won't have been
> restored but thankfully it's backed up nightly and doesn't change too
> often :-) )
Well, I'd be a little worried about whether the base backup was
self-consistent, but if it was taken at a time where the DB was idle
then you can probably get away with this.
> What I'll probably do is try to simulate the same process again on a
> different machine to get myself a bit more familiar. Is there any other
> situations you can think of where this may also be relevant, or is it
> just when dropping a complete database?
AFAIK the only operations that have non-rollbackable side effects are
CREATE/DROP DATABASE and CREATE/DROP TABLESPACE. For any of these,
you'd end up with inconsistent state if you try to stop replay just
before the commit record.
regards, tom lane