[OpenAFS-devel] Migrating Fedora Build Slaves -- which to keep?

[Benjamin Kaduk]

On Sun, Nov 06, 2016 at 09:08:36PM +0100, Stephan Wiesand wrote:
> Hi Derek,
> 
> On Nov 5, 2016, at 20:38 , Derek Atkins wrote:
> 
> > I guess the question is: for how long after a release is EOLed should we
> > keep the build server around?  Is there a general OpenAFS build server
> > policy for how long to keep a build slave around?  If not, should we
> > have a policy?
> 
> 
> I doubt that there's a policy. From the point of view of someone pressing the submit button based on code review and buildbot verification: the more the marrier, as long as a build slave is reliable and not terribly slow.
> 
> I don't think we should spend effort on supporting obsolete distributions. But we sure don't deliberately break them. And every successful build on a different distro, with a different tool chain and against a different kernel is an important data point for judging the correctness of a change.
> 
> If keeping an old build slave alive keeps you from developing/submitting changes, reviewing other's, testing prereleases etcetc, that would be a good reason to cull it and rather spend the time on those other activities. Otherwise, well, please see above.

I agree with Stephan's points.  In terms of triaging which VMs to keep (if there
is a need to cull), we definitely get benefit from the bleeding-edge systems,
to pick up new compilers and their compiler warnings (as well as their new
kernels, though hopefully we will be on top of that already).  There is also value
in the old(est) sytems as a sort of regression test, though I don't think I
have any examples where those would actually catch anything, so maybe that's
just a red herring.

> Whatever you decide to keep or cull, thanks a lot for providing those build slaves all along!

Seconded!

-Ben