Re: [atomic] OpenStack Magnum concerns about Project Atomic

[Steve Gordon <sgordon redhat com>]

----- Original Message -----
> From: "Matt Micene" <nzwulfin gmail com>
> To: "Josh Berkus" <jberkus redhat com>
> 
> > * Depending on timing age of components in Fedora Atomic leads to a need
> > for Magnum folks to rebuild image, at least until such time as more of the
> > pieces are containerized (k8s/etcd/flanel) [1].
> >
> 
> CoreOS is a standalone distro, and can move at whatever speed for whatever
> component they like.  Our packages move at the speed of Fedora.  The Atomic
> members may be maintainers of a component, but if Fedora decides that, for
> example, docker 1.10 is a breaking change that needs to wait until F24,
> there's not a lot we can do about it for a F23 based Atomic Host.

Certainly, the issue for the Magnum folks though is that until the services noted (kubernetes/etc/flannel) are containerized then the path they are pushed down to replace them is to rebuild the image from scratch which comes with its own pitfalls. Interestingly I don't believe they have actually been as concerned about the docker version in use thus far.

> * Doesn't really feel like an active global community versus e.g. CoreOS.
> 
> Yep, I agree.  And fairly small overall.  Part of why docs and finding
> answers online are a problem.  Depending on the problem, there's still the
> possibility of needing to find 1 specific community member to get an
> answer.  This is also why things like containerized components are just
> open PRs at this point.  (IMO)

There does also seem to be an issue of it being kind of spread out, even when posting this I wasn't sure if it belongs here or on the Fedora cloud list (and of course there are equivalent IRC channels for each too), and I just today learned there is an ask.atomicproject.io as well! Not sure what if anything can be done about this.

> the Magnum folks are primarily using the Fedora Atomic images as their base
> > but I was recently pointed to this IRC discussion where the Magnum folks
> > outlined some of their concerns resulting from interacting with the Atomic
> > images which are causing them to look elsewhere:
> 
> 
> I don't think anyone is specifically ignoring downstream use, but I wasn't
> aware that there *was* any downstream use.  Based on:
> 
> https://lists.projectatomic.io/projectatomic-archives/atomic-devel/2015-April/msg00075.html
> 
> I didn't know Magnum was back to trying Fedora Atomic.  Maybe someone else
> in the channel or on the list has some knowledge of downstreams that we
> need to make people aware of?
> 
> - Matt M

The proposal to instead use a normal Fedora cloud image is mainly a result of the fact that the tools for customizing such an image are much better understood and documented. At this time though I believe the image used in the CI gate for Magnum is still one of these customized Fedora Atomic Host 21 images:

    https://fedorapeople.org/groups/magnum/

There is a proposal up to make the required tweaks to Magnum to attempt to use Fedora Atomic Host 23 image:

    https://review.openstack.org/#/c/276232/

Timing wise the reality is though that even if the above is merged for now I imagine they will eventually have a need/want to update one of kubernetes/flannel/etcd before F24 timeframe bringing them back around to the question about containerization of these pieces :).

-Steve

> On Wed, Feb 24, 2016 at 10:44 AM, Josh Berkus <jberkus redhat com> wrote:
> 
> > On 02/24/2016 07:13 AM, Steve Gordon wrote:
> >
> >> Hi all,
> >>
> >> Not sure if this is the right place as the Magnum folks are primarily
> >> using the Fedora Atomic images as their base but I was recently pointed to
> >> this IRC discussion where the Magnum folks outlined some of their concerns
> >> resulting from interacting with the Atomic images which are causing them
> >> to
> >> look elsewhere:
> >>
> >
> > Seems like as good a place as any.
> >
> >
> >> http://eavesdrop.openstack.org/irclogs/%23openstack-containers/%23openstack-containers.2016-02-15.log.html
> >>
> >> This is in some ways a continuation of this earlier thread:
> >>
> >>
> >> http://lists.openstack.org/pipermail/openstack-dev/2016-February/thread.html#85699
> >>
> >> Trying to boil down the comments into a summary:
> >>
> >> * Depending on timing age of components in Fedora Atomic leads to a need
> >> for Magnum folks to rebuild image, at least until such time as more of the
> >> pieces are containerized (k8s/etcd/flanel) [1].
> >>
> >
> > One thing I don't get from the chat is examples.  Are they really building
> > this with a release of Kubernetes which is < 2 weeks old?  Or are our
> > packages older than that?
> >
> > * Rebuild image documentation lacks troubleshooting information (I
> >> encountered this myself, there are plenty of places for things to go off
> >> the rails in the compose and create-vm-disk steps and the tooling provides
> >> very limited error output in some cases).
> >>
> >
> > Yeah, docs are a problem.
> >
> > * Lack of clarity around if or when originally proposed build cadences
> >> will be hit for Fedora Atomic - I think this was originally proposed as
> >> 2-weekly but it's hard to tell if that has actually been hit to an
> >> outsider. To me it looks like yes [2] but do we record anywhere when a new
> >> build was pushed out to the mirrors as the current stable?
> >> * Doesn't really feel like an active global community versus e.g. CoreOS.
> >> I know this is a challenging thing to "solve" but I list it as it is part
> >> of their concern, I myself have asked a question in the #atomic IRC and it
> >> was eventually answered but only when US East coast folks were up and
> >> awake
> >> again.
> >>
> >
> > This seems to center around availability on IRC.  Part of the problem
> > there is that folks are spread out across 5-6 different channels
> > (#fedora-cloud, #nulecule, etc.).  I'll try to get people to remember to
> > also log into #atomic.
> >
> > I know this is very high level feedback in many cases and difficult to
> >> action but wanted to highlight it . Scott has already answered my query on
> >> the Fedora cloud list w.r.t. containerization plans for the services
> >> mentioned which I think would help Magnum folks a lot [1], and I see we
> >> are
> >> apparently now doing two-weekly updates to the Fedora cloud image [2]
> >> though as I mentioned it would be great if we can illustrate the history
> >> of
> >> this in terms of regular builds produced/pushed.
> >>
> >
> > I think it's pretty valid feedback and I'm glad you're relaying it to us.
> >
> >
> > --
> > --
> > Josh Berkus
> > Project Atomic
> > Red Hat OSAS
> >
> >
> 

-- 
Steve Gordon,
Sr. Technical Product Manager,
Red Hat OpenStack Platform