On Thu, Nov 02, 2017 at 04:54:38PM +0100, Pavel Hrdina wrote:
On Thu, Nov 02, 2017 at 03:42:56PM +0000, Daniel P. Berrange wrote:
> On Thu, Nov 02, 2017 at 01:45:51PM +0100, Pavel Hrdina wrote:
> > This fixes libosinfo RPM build, we need to set "OSINFO_SYSTEM_DIR"
for
> > RPM build as well since the spec file runs tests and they need valid
> > osinfo-db. Broken by <359b069f7e4700bee337588eab0a9d83cac94010>.
>
> I'm not suggesting you fix this right now, but our current RPM build
> process is very broken by design, because we are completely ignoring
> dependancies between packages. We've hacked around this by culling the
> BuildRequires lines from the spec file before running rpmbuild, and
> relying on a bit of luck to have the rpmbuild find the stuff we just
> built into $VIRT_PREFIX. As this test suite problem shows though,
> this is very fragile.
I completely agree with it, it's fragile and it uses all the environment
variables to make sure that all the dependencies can be resolved.
> One way to fix this is to create a yum repo populated with all RPMs
> we build during CI. Then setup a mock vroot to do the builds in,
> which includes this local yum repo. That way we can honour the RPM
> deps correctly getting a more accurate test of RPM build process.
>
> My main concern with this is that mock is kind of slow to bootstrap
> the initial vroot, so this would slow our CI somewhat.
We need to look into it how much it will slow down our CI since we have
very limited resources. Maybe it's time to ask whether we can have more
resources.
I think it is well overdue for us to acquire more hardware for Jenkins
slaves. If CentOS have a 2nd machine they can spare for us then great,
otherwise we should try to push RH to fund some...
Regards,
Daniel
--
|:
https://berrange.com -o-
https://www.flickr.com/photos/dberrange :|
|:
https://libvirt.org -o-
https://fstop138.berrange.com :|
|:
https://entangle-photo.org -o-
https://www.instagram.com/dberrange :|