Glen Parker <glenebob@nwlink.com> writes:
>> Tarring hot database files still gives me the willies. But then, I
>> wear belt and suspenders.
> I understand. A list of "file changed while we read it" errors is just
> a little unnerving at first!
> I did quite a few end to end backup/PITR tests, and no matter what I did
> to the DB during backup, no matter how many errors tar or cpio produced,
> I have been unable to find any problems with the end result.
In theory, at least, this is exactly as reliable as Postgres' crash
recovery: it's essentially the same mechanism that gets us back to a
consistent state on-disk after a crash. So don't worry too much about
all those gripes from tar. (The only real problem with 'em is they
might distract you from a more serious problem, like out-of-space.)
regards, tom lane