[Pkg-postgresql-public] Bug#764705: Bug#764705: Bug#764705: Bug#764705: Bug#764705: postgresql-9.4: ERROR: The database format changed between beta 2 and 3. Please dump, but how?

[Christoph Berg]

Re: Martin Pitt 2014-10-12 <20141012105918.GD3788 at piware.de>
> Peter Eisentraut [2014-10-11 17:48 -0400]:
> > I think the mistake here was to upload postgresql-9.4 into unstable
> > before it was released.
> 
> I agree. We should have kept it in experimental until the release
> candidate and prepare the extensions there, like in earlier releases.

Yeah, though it was pretty unexpected that there was another catalog
version bump after beta2 :-(. The general plan was still ok in my
view.

> This is a rather awkward position for upgraders to be in, as we
> essentially cannot automate this robustly. This is exactly the reason
> why we introduced the p-common infrastructure more than a decade ago,
> as trying to do the dumping/restoring in maintainer scripts led to an
> eternal stream of upgrade failures and some data loss. But
> pg_upgradecluster can't help us with this situation, as we cannot
> install beta 2 and 3 in parallel.

There's probably ways to script that, but they invoke copying the old
server files to a temp location in the preinst, and then use that
later. Which leads us back to the old /usr/lib/postgresql/oldwhatnot
setup that was used before the introduction of pg-common...

I don't think it we can make that construction work without the risk
of it exploding in one way or the other, so the approach to let users
(pg_)dump and restore the clusters is 

> So I think the best we can do here is to document how to do the
> upgrade, and then not repeat the premature release of betas to
> unstable?

The error message should be replaced by a nicer debconf note. I'll
look into that later this week. (In the hope the debconf works in the
preinst...)

Re betas, it really had better already already been the RC phase to do
that, yes. Though I don't think the very timing with the Debian freeze
would happen again, so it'll be less of a problem next time.

Christoph
-- 
cb at df7cb.de | http://www.df7cb.de/