[rdo-dev] Packaging issues about upcoming changes in Sahara

[Alfredo Moralejo Alonso]

On Mon, Dec 3, 2018 at 2:23 PM Luigi Toscano <ltoscano at redhat.com> wrote:

> On Wednesday, 28 November 2018 18:43:04 CET Alfredo Moralejo Alonso wrote:
> > On Tue, Nov 27, 2018 at 4:02 PM Luigi Toscano <ltoscano at redhat.com>
> wrote:
>
> > > The source package openstack-sahara generates the following packages:
> > > - python{2,3}-sahara contains the shared python libraries (with tests
> > > ending
> > > up in python2-sahara-tests). Almost all the other subpackages depends
> on
> > > it.
> > > - openstack-sahara-image-pack is not a service, but a standalone
> command
> > > used
> > > to build images. It only depends on python{2,3}-sahara.
> > > - openstack-sahara-common contains few shared components and programs
> > > shared
> > > by all services.
> > > - openstack-sahara-api and openstack-sahara-engine depend on
> > > openstack-sahara-
> > > common and contain only the unit file and the entry point for the
> services
> > > of
> > > the same name.
> > > - openstack-sahara depends on -api,-engine,-common,-image-pack, and
> > > despite
> > > the name, contains the long-deprecated openstack-sahara-all unit file
> and
> > > entry point. Historical reasons (it was originally the package holding
> the
> > > sahara service).
> > > - there is also openstack-sahara-doc for documentation.
> > >
> > > == The main disruptive change is the spliting of plugins from the
> Sahara
> > > codebase. They are being moved in their own repositories:
> > >
> > >
> http://specs.openstack.org/openstack/sahara-specs/specs/rocky/plugins-outs
> > > ide-sahara-core.html
> > >
> > > Each plugin will get its own package, with subpackages for the main
> code,
> > > for
> > > the tests and the documentation. Please note that each plugin package
> > > depends
> > > on python{2,3}-sahara, while the core does not depend on plugins.
> > >
> > > Sahara services can still start even if no additional plugins, thanks
> to
> > > the
> > > fake plugin still living inside sahara.git.
> > >
> > > Now, I was wondering if there is a way to automatically installing the
> > > known
> > > plugin subpackages when installing or upgrading (with dnf/yum)
> "sahara",
> > > without introducing circular dependencies. Does it make sense to do it?
> > > Maybe
> > > introducing a new source package which only depends on all sahara
> > > packages,
> > > services and plugins? Would it make sense to do it?
> > > If it makes sense, would it make sense to rename
> openstack-sahara.noarch
> > > as
> > > openstack-sahara-core.noarch, and move the openstack-sahara "catch-all"
> > > subpackage to this new source package?
> >
> > In that case, the new package would only depend on plugins or also in
> > services and common?
> On all of them. The idea would be to keep binary openstack-sahara package
> as
> the package which installs everything, as it is the case right now.
>
> >
> > I think this is similar to what we have with tempest and separated
> plugins.
> > For this case we have a subpackage tempest-all that depends in the
> plugins.
> > To allow bootstraping the repo avoiding circular dependencies we use a
> > parameter repo_bootstrap that we set to 1 when boostraping a repo [1]. I
> > think you could do something similar and add all plugins as requires for
> > the desired packages, i.e. openstack-sahara-plugins if you want to have a
> > subpackages only installing plugins or even as requirements of -common if
> > what you want is to always get plugins installed for each service
> > installation.
>
> Thanks for the hint. So here we could use this trick (toggle the value of
> the
> parameter after the plugins packages are available) to make the binary
> openstack-sahara package depend on all plugin packages. I don't think that
> there is a need for a binary openstack-sahara-plugins package.
>
> The drawback of this approach is that packagers need to remember to do it
> at
> each cycle! :)
>
>
Yes, maintainers need to toggle the parameter to off and on in bootstraping
new releases.


> On the other side, the only alternative (new source package) may have a
> higher
> initial cost - or maybe not. Is there any specific reason why you didn't
> follow this path with openstack-tempest-all?
>
>
So IIUC, what you propose is to move openstack-tempest-all to a new package
(srpm) which would only have a subpackage with requires to all Sahara ones,
right?.

I guess It should work but we would probably hit problems with tooling for
the new package because it wouldn't have an associate source repo. We
probably only need to create some fake repo. I like more the idea of using
the repo_bootstrap for consistency (it's something we use in different
packages with similar issues).


> >
> > [1]
> >
> https://github.com/rdo-packages/tempest-distgit/blob/rpm-master/openstack-te
> > mpest.spec#L131-L171
> > > == Smaller change: the openstack-sahara-all service is going away,
> maybe
> > > in
> > > stein, but for sure in train. Would it make sense to keep
> openstack-sahara
> > > (or
> > > openstack-sahara-core, see the previous point) as empty package just
> for
> > > dependency reasons?
> >
> > Yes, I think it makes sense to have a metapackage to install all
> supported
> > plugins.
>
> Ack, thanks.
>
> Ciao
> --
> Luigi
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rdoproject.org/pipermail/dev/attachments/20181204/eb2e2034/attachment.html>