On Wed, 2010-07-21 at 10:53 -0400, Kamil Paral wrote:
----- "James Laska" <jlaska(a)redhat.com> wrote:
> On Wed, 2010-07-21 at 10:06 -0400, Josef Skladanka wrote:
> > Yes, technically, it is breaking the upgrade path, the question was,
>
> > if we want to support users who basically broke the path for
> themselves
> > by removing one of the repos before updating.
>
> This test is likely to be initiated by the post-bodhi-update watcher
> right? That watcher provides information about what target the
> update
> is for. Can we keep it simple and assume for now ...
>
> * if the update is intended for a *-updates-testing repository,
> include all -updates-testing tags
> * if the update is intended for *-updates repository, include
> only
> *-updates tags
>
> This seems like it would capture the most common scenarios and we can
> handle exception scenarios at later date (or as an informational
> result)?
I'll answer for Josef, because he won't return till Friday.
The basic scenario:
F12 stable -> F13 stable
is crystal clear. No upgrade path problems must occur.
With updates-testing however, issues appear. One of the issues was described
by Josef pretty comprehensibly. Another issue is that updates may be
withdrawn from updates-testing. So you can push the same version to F12
and F13, and then withdraw the one from F13. Kaboom. AutoQA is not able to
catch this, the withdrawal happens only after primary AutoQA check is
finished.
So the question basically is - how complete do we want to have our upgrade
path test? It can check just the common stuff, as James just suggested. Or
we can try to cover all the corner cases - but some of them will probably
need some strict policies to take place - like the one described by Josef
(pushing to F12 updates-testing only after it has landed in F13 updates).
It seems to me that the easiest solution is to make AutoQA check just the
common scenarios and offload some of the corner cases to Bodhi, because
it can prevent some things much more easily than we can (like the
aforementioned package withdrawal). Of course that means the logic of this
test is then split before AutoQA and Bodhi, it's not in a single place.
Topic for another time ... but you raise another discussion. Is bodhi
package withdrawal a new test event watcher?
Thanks,
James