On Mon, 3 Aug 2015 16:17:18 -0600
Tim Flink <tflink(a)redhat.com> wrote:
...snip...
> * bodhi2 isn't yet packaged in Fedora (it and it's deps
are in a
> copr):
http://copr.fedoraproject.org/coprs/lmacken/bodhi2/
I didn't think that there had been much testing of the APIs yet. Last
I heard, the API wasn't ready for testing - has that changed?
I think it is now, but hopefully Luke can answer.
The api should be handled mostly in python-fedora, so not sure how
many direct api users would be affected.
> So, to get that last bit we need:
>
> * API users. I thought we had a list somewhere, but I cannot seem to
> find it. At least fedora-easy-karma and fedora-gooey-karma use it.
> Are all of these using python-fedora for the api? Do we need to
> update that and bodhi-client?
libtaskotron, resultsdb and blockerbugs use the Bodhi API through
python-fedora. None of them have been tested against bodhi2.
Right. Luke is working on python-fedora now, hopefully soon we can test
them.
> * We need bodhi2 in fedora/epel. Were we going to do it as a
> seperate package, or just an update to the existing 'bodhi'
> package. I assume maintainers would need the new client to do
> 'fedpkg update' ?
>
> * Do we need any fedpkg changes? Or anything else? How do we want to
> handle the switchover?
>
> * Is taskotron all ready to go with bodhi2? or more changes there?
Not really. Preparing for bodhi2 has been on the back burner for us
since it didn't sound like production was going to be ready before F23
was released.
I'll try to take a look into what's left but the guy who has been
doing most of that work is on PTO until Flock.
Hopefully there will be very little to nothing for it to need to do
(aside from using the new python-fedora).
> * We will need an outage to switch over. I assume the
bodhi1->bodhi2
> db conversion shouldn't take too long? But once we convert and make
> live we are commmited.
>
> * Anything else I missed?
>
> I'd like to propose we schedule the switchover/outage for
> 2015-08-19. (Provided alpha doesn't slip). This will allow us a few
> days to recover from flock and get anything lined up we need to.
How heavily has bodhi2 been tested by folks who aren't bodhi2
developers? The idea of swapping out bodhi mid-release makes me a
little nervous since hiccups could affect the release process.
Well, it could affect the release process no matter when it's done.
It could affect the flow of updates anytime (always some of which are
important, we have no way to tell everyone to not have major bugs or
security issues for a while. ;)
I realize there are ~3 weeks between the proposed date and F23 beta
freeze and that there are going to be bumps in the road no matter
when the switchover happens. However, the tester in me is thinking
more along the lines of what might go wrong in the near future if the
more corner-casey stuff (freeze etc.) is hit so soon after deployment,
especially if there's a bug that's taken down the frontend for at
least a couple hours in addition to needing a database
wipe/repopulate 2 weeks before the proposed go-live date.
Well, frankly the longer we wait the worse things are going to get.
The problem is that while we keep using bodhi1, Luke has to bandaid and
poke at it to keep it going with our needs, which means he's NOT
working on bodhi2, which means we run further behind.
We failed completely to 'release early and often' with bodhi2, but
really IMHO we need to bite the bullet and get it done now that we are
close.
If we find anything that is obviously needing fixing before we roll it
out, we can delay, but I do not want to wait until after f23.
kevin