[debian-mysql] mysql-defaults into unstable?

[Robie Basak]

On Thu, Jul 21, 2016 at 10:38:46PM +0300, Otto Kekäläinen wrote:
> > Do we want to put mysql-5.7 in at the same time? Because otherwise
> > src:mysql-5.6 still produces mysql-common, which is not ideal.
> 
> There is no strict dependency there. The only thing we need to be
> careful about is not to upload any new versions of mysql-5.6 anymore
> (or at least current git master that would provide mysql-common that
> conflicts with the new mysql-common).
> 
> Let's just upload mysql-5.7 to unstable when it is ready, it does not
> need to be in sync with mysql-defaults.

OK.

> > What do you want to do in Ubuntu? Should we defer the move to
> > mysql-defaults until Yakkety+1? If not, we need to decide what we're
> > doing very soon, and if so, I'd like the opportunity to blacklist
> > mysql-defaults and mariadb-10.0 so they don't auto-sync, so we can
> > migrate in Ubuntu when we're ready.
> 
> We cannot really blacklist mysql-defaults, as any of the other
> packages in the universe might declare it as a dependency. All you can
> do is provide a customized version that behaves differently if you
> need so. As Yakkety is not an LTS version I would maybe not hesitate
> putting in the latest software versions there. The less there is
> delta, the less we have to maintain, right?

Yakkety's feature freeze is August 18. If it were August 18 today, then
there would be no question: we'd stop autosyncing from Debian and
mysql-defaults would wait until next cycle. I wouldn't want a feature
freeze exception for this change because I don't think there would be
any immediate benefit for Ubuntu users.

So given that Ubuntu's door essentially closes four weeks from today, is
it the right time?

> as any of the other packages in the universe might declare it as a
> dependency

It seems unlikely to me that much would be affected in four weeks,
especially with the timetable you've set out below. If a package did get
updated and synced to Ubuntu in that time, it would be blocked from
entering the release pocket on the basis that it is not installable
(like unstable -> testing). Except that I think it would still be
installable, because the alternative would be used.

I agree this isn't ideal, but I also don't want to spend time after
feature freeze sorting things out. If we did it early in Yakkety+1, then
we'd have more time.

I'm not yet decided on which approach is best, and welcome further
discussion.

> Summary:
> - Uploading mysql-defaults will provide a new mysql-common but which
> has exactly the same contents as it was already in mysql-5.6/5.7.
> Regressions are unlikely. The other things mysql-defaults provides
> will not have any effect until other packages start to use them. So
> this should be safe to upload at any time. We are doing a slow
> roll-out just in case.
> - Uploading mariadb-10.0 will provide new libmariadbclient18 packages.
> These again have no effect until other packages start to use them.
> This should be safe to upload too.
> - Uploading mysql-5.7 will provide new mysql-* packages. This will
> cause new installs of mysql-server to behave differently. This has
> however been tested in Ubuntu Xenial and Yakkety already, and Robie is
> working on additional fixes. With those fixes it should be safe to
> upload too.

Agreed.

> My suggestion:
> - upload mariadb-10.0 now
> - upload mysql-5.7 next week
> - upload mysql-defaults the week after that

I think this plan will be fine for Debian. Thank you for putting it
forward.

I'm still not sure what to do in Ubuntu, but we can figure it out based
on this timetable for Debian.

Thanks,

Robie