Open Stack

On 01/06/2016 07:02 AM, Jeremy Stanley wrote:
> Also, that was for being able to test things which only run on
> Fedora. Since "testing Debian" is not the _stated_ goal (as far as I
> know), there are other incremental ways to get started designing and
> running these package build jobs on images for platforms we already
> have working today: i.e. (c)debootstrap that Debian Sid chroot on
> Ubuntu Trusty, install a pbuilder or sbuild DEB from a Debian
> mirror, etc.
> 
> I've gotten very tired of pointing out that Thomas has come at this
> under the assumption that he must start with a Debian-based system
> to host these jobs, but if they're going to generate/update/archive
> a chroot anyway then it seems less critical that someone solves
> getting Debian worker images built by Nodepool before he can make
> progress on what he actually wants to accomplish. In fact, our
> worker image builds (for Fedora, CentOS, whatever) happen in
> distro-specific chroots created on an Ubuntu Trusty server anyway...
> three cheers for diskimage-builder!
> 
> Don't get me wrong, I'd love to see someone maintaining Debian
> images in Nodepool for us too, but Thomas _could_ be working on
> writing and testing jobs today that are "close enough" instead of
> complaining about how nobody has time to build an ideal environment
> for him before he can even get started.

I'm not "complaining". This sounds like a kid whining, which isn't what
I'm doing. I've explained the facts which made me stuck where I am, and
what needs to be done to go ahead, hoping to get help.

Plus I don't need someone to do it all, just the beginning of the patch
to help me understand what needs to be done, at which point I hope to
take over.

Once such image is there, I'll maintain it the best I can. I'm already
maintaining the official Debian OpenStack images available at
http://cdimage.debian.org/cdimage/openstack/, which IMO could be used as
a base image (provided we can arrange to fix the cloud-init missing
systemd service files, which isn't hard to do IMO).

Now, about using Ubuntu instead of Debian. We've gone through this
multiple times. Since you are going back on this, let me try again to
explain my view and convince you that what you're proposing isn't a
viable option, even for the short therm.

Just using a chroot is *not* an option, because I want to use Debian to
do tempest runs. This isn't something just to be done in the future, but
right now, as I already have all the scripts available to make it
happen. This really is needed to do initial gating for the packaging.
Using Ubuntu isn't an option here, especially that it uses upstart
rather than systemd, so we'd be checking very different things, and
also, I'm not even sure how to hack the VM to make it boot on the chroot
rather than on the original OS (and rebooting the VM is needed when
setting-up the network).

There's also stuff which are different between Trusty and Jessie. For
example. dpkg-parsechangelog -S doesn't work on Trusty, and this is
needed *before* getting into the sbuild chroot (to find out about
package name, version, etc. before building). This is just one example,
there are others, which all lead me to say it's mandatory to build
packages in the target host OS, even if we're using sbuild which
provides a chroot.

Anyway, just using Trusty to package in Debian feels at so many levels
like the wrong thing to do which will for sure lead to a lot of loss of
time. I don't want to waste such a time.

Moving forward, I created a patch, but really, it feels like I don't
know what I'm doing and many things are missing. Comments and
improvement are of course more than welcome. [2]

Cheers,

Thomas Goirand (zigo)

[2] https://review.openstack.org/264726