On Tue, Oct 6, 2015 at 11:21 PM, Bohuslav Kabrda <slavek(a)redhat.com> wrote:
----- Original Message -----
> On Fri, Oct 2, 2015 at 4:01 AM, Bohuslav Kabrda <slavek(a)redhat.com> wrote:
> > ----- Original Message -----
> >> On Thu, Oct 01, 2015 at 10:00:43AM -0500, Scott Collier wrote:
> >> > >My opinion is that the dist-git-for-dockerfiles plan should make
> >> > >Dockerfiles obsolete.
> >> > How would users who aren't necessarily involved in the Fedora
> >> > process experiment building images with a Fedora base? Right now
> >> Hopefully this new dist-git can be fronted by pagure, so it'd be a
> >> matter of visiting a web site like https://pagure.io/fedora-bootstrap
> >> (not a docker example, just a random one) and either downloading the
> >> docker file or doing a git clone.
> > Heh, it seems that my career as Fedora-Dockerfiles comaintainer may be
> > rather short :)
> > I think having a web frontend with pull requests for the new dist-git is an
> > awesome idea. I'm +0.9 for pagure. The advantage is that it will be
> > completely under Fedora control, the small downside is that potential
> > contributors from outside Fedora will have to create Fedora account, which
> > might scare some people off.
> From what I understand is that pagure uses OAuth and at the moment the
> only OAuth plugin in use is for FAS, we might be able to request to
> enable some short list of federated logins if:
> 1) that is in fact possible
> 2) the authors of pagure are open to it
> 3) the Fedora Infrastructure team is OK with it
That would be nice! I'll try to ask pingou whether this would make sense to him
(hopefully in the start of next week, I'm going to a conference for the rest of this
+1 - Sounds good.
> >> > there's a rpm created from fedora-dockerfiles that includes all
> >> > Dockerfiles and makes it easy to experiment by placing all the
> >> > Dockerfiles in /usr/share/fedora-dockerfiles. If we keep them both,
> >> > it's somewhat duplicate work. I'm just curious how it would
> >> > Right now the barrier to entry for experimentation is low. I'm
> >> > concerned about raising that.
> >> Yeah, that's a good concern. We could also perhaps automate pulling all
> >> the dockerfiles from the dist-git into fedora-dockerfiles and keep
> >> that, for people who want to work with it that way.
> > So IIUC, the standard way to get dockerfiles from dist-git would be
> > clone mariadb-docker" or similar. Perhaps we could provide a wrapper that
> > anyone, even without Fedora account, like "fedora-get-dockerfile
> > or "fedora-get-dockerfile mariadb" (this would invoke "fedpkg
> > --anonymous mariadb-docker").
> Ultimately people could just git clone from the git repo in pagure.
Yeah, you're right.
> The only real thing that makes DistGit "special" is the branch layout,
> relationship, and some git hooks. Otherwise it's still just git and we
> can aggregate that information any way we choose. If there's
> ultimately a desire for a web hub, it's possible we could get
> something together similar to Fedora Packages
> >> > Also, the plan was for Vasek to submit Nulecule PRs to
> >> > Fedora-dockerfiles, at some point, so people could experiment with
> >> > them as well. I'd also like to see k8s example json / yaml files
> >> > associated with select fedora-dockerfiles for easy experimentation.
> >> > Would the nulecule / k8s get pushed into the dist-git as well?
> >> Maybe? Would it make sense for these to go together with the
> >> dockerfiles they're associated with in a git repo at that level, or
> >> would they be stand-alone and reference other repos? (Do you have some
> >> concrete examples?)
> > So I think that kubernetes/Nulecule examples should be standalone, since
> > most often they'll reference multiple images. What I mean is that they
> > would be a good fit for the current fedora-dockerfiles repo, but if we
> > split the repo into multiple dist-git repos, they won't fit in any one of
> > these.
> I'm not sure that I follow the motivation behind keeping them
> standalone, could you elaborate why? Also, when you say "reference
> multiple images" so you mean as a Nulecule spec or $other/$similar?
So *assuming* we split fedora-dockerfiles in one-dockerfile-per-dist-git-repo style and
there's a kubernetes/Nulecule file that references (*) more of them, it just
doesn't fit into any one of these one-dockerfile repos.
(*) It's pretty much as you said it, an application based on e.g. a kubernetes config
file will *typically* operate with (and thus reference) more than just one image.
Yes, I agree. For things like kubernetes and/or nulecule application
definitions, I think those should live somewhere external
independently at least for now. This isn't something we've currently
been planning to handle in the Layered Image Build System, at least
not at first. I'm not even sure what the implications are from a build
system perspective because I was under the impression that those were
just runtime definitions, but as time goes on this can definitely
> >> Another approach to all this would be to keep all the dockerfiles in
> >> one place -- it just means that everyone who has any commit access to
> >> anything gets commit access to everything, which we might not want as
> >> we scale.
> > Yeah, scaling is a valid point. I think splitting into multiple repos is a
> > good way to go assuming we address the concerns that Scott has raised.
> I think the main concern here (unless I"ve misunderstood) is
> visibility and collaboration which I think we can solve. I like the
> idea of the different Dockerfiles being separate similar to packages
> in DistGit purely because that allows people with expertise in certain
> areas to participate in the areas they care about and avoid all the
> other line noise of a shared repository.
> Just my $0.02
>  - https://apps.fedoraproject.org/packages/
> >> --
> >> Matthew Miller
> >> <mattdm(a)fedoraproject.org>
> >> Fedora Project Leader
> > --
> > Regards,
> > Slavek Kabrda
cloud mailing list
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct