On Fri 2021-05-07 09:43 , Florian Weimer wrote:
> * Peter Robinson:
> > On Fri, May 7, 2021 at 7:58 AM Florian Weimer <fweimer(a)redhat.com>
> >> We want to use kernel rebuilds as a gating test for toolchain updates.
> >> Unfortunately, per
> >> Information for package kernel
> >> <https://koji.fedoraproject.org/koji/packageinfo?packageID=8>
> >> the last successful rawhide kernel build was on 2021-04-28.
> >> The question is what we should do about build failures like this.
> >> Should we use non-rawhide kernels for our gating tests?
> > Well rawhide is generally OK, but it's not unusual for kernels to fail
> > during the two week upstream merge window, I've seen all arches in x86
> > cause issues during the merge window. Outside of that things tend to
> > be OK.
> Hmm. Currently our gating check simply does
> koji build --scratch --wait --fail-fast f34-build-side-40913
> (Well, it's a bug that we use the rawhide kernel branch on the f34
> branch, but we'd still be impacted for actual rawhide builds of
> toolchain packages, of course.)
> This koji command is nice and short, but it assumes that the dist-git
> branch is actually buildable.
> Should we use a SRPM from a compose instead? But composes can be really
> old, and if we need a kernel fix to get things building again with newer
> toolchains, then this could potentially block toolchain upgrades for a
> long time.
If I fedpkg clone the kernel repo (and switch to the appropriate
branch) ... is there a way to identify (if not 100%, then at
least with high probability) the latest commit that should be
usable for the rebuild test?
You could query koji if the last build failed or not, and if it failed
how long ago that was. If it built OK in the last 24 hrs it's likely a
good indicator it's fine. The failures are mostly during the merge
window, so 2 weeks out of 10. You could also check if there's a
failure and if it's got rc0 in the NVR.