Re: pg_upgrade failing from 9.3 to 9.4 because "template0" already exists

Stephen Frost <sfrost(at)snowman(dot)net> writes:
> * Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
>> Perhaps pg_upgrade should deliberately ignore template0 regardless of
>> datallowconn? And/or we should hard-wire that into pg_dumpall?

> My thinking would be that pg_dumpall should be hard-wired for template0
> (just like it is for template1..) and that we should *not* be excluding
> databases that are marked as datallowconn = false.. That said, it's not
> clear to me what to do there instead. Maybe throw an error or a
> warning? The point of pg_dumpall is to dump *all* the databases and at
> least the manpage doesn't appear to say anything about "but ignores
> databases with datallowconn = false".

I think pg_upgrade and pg_dumpall may be two different use-cases.
pg_upgrade should definitely throw a hard error if there are any
non-template0 databases that it can't connect to, because the alternative
is losing such databases during the upgrade. I'm not sure that the
argument is so black-and-white for pg_dumpall, though. Nobody's ever
complained about it skipping unconnectable databases, and that behavior
has been there since we invented datallowconn (cf commit 2cf48ca04bf599).

regards, tom lane