On Sat, Jan 4, 2025 at 5:08 AM Paul Moore <paul@xxxxxxxxxxxxxx> wrote: > > On Fri, Jan 3, 2025 at 8:48 AM Ondrej Mosnacek <omosnace@xxxxxxxxxx> wrote: > > > > There seems to be another issue with the aarch64 builds now: the > > latest -rc3 RPMs are not present in the repo, even though they are > > referenced in the repo metadata ... > > Please let me know if you see this happen again, but I think this was > just an usual combination of things coming together, including some > rather spectacular timing to catch the repo at just the right point in > time to see this. I'll explain a bit more below if you're curious, > but everything appears to be working correctly with the 6.13-rc5 > builds, at least on my test systems: > > * https://groups.google.com/g/kernel-secnext/c/i4UAqrY5E8o > > As far as to why this happened with the -rc3 build, let me first > provide some background: > > - In an effort to limit the amount of disk space needed for > repo.paul-moore.com I only keep the last 14 days of builds on that > system. > > - The job which uploads the builds to repo.paul-moore.com first > removes all packages built more than 14 days ago, then uploads the new > builds, and finally regenerates the repo metadata using createrepo. > While there is a window where the packages have been removed and the > metadata has not yet been updated, this generally isn't a problem > because 1) the time window is relatively short and 2) there generally > isn't much interest in "old" secnext kernel builds (it is somewhat > counter to the whole bleeding edge testing idea). > > - When there is a significant backlog of packages to build, as was the > case when I restored the aarch64 builder, the package build job starts > with the newest src.rpm first and works backwards, because once again, > people generally only care about the most recent secnext builds so > this approach gets that build out quicker. In this particular case > that was likely 6.13-rc3. > > So, despite 6.13-rc3 being the "latest" kernel build in terms of > version, it was one of the older packages in terms of build date and > thus when the repo.paul-moore.com was updated with fresh builds, the > "old" 6.13-rc3 packages were removed. If the testing farm tests > happened to start during that window where the packages had been > removed, but the metadata not yet updated, I suspect you would have > run into the problem you describe. Yes, when I re-ran the jobs later, the installation succeeded (and it installed an -rc5 kernel). Thanks for the explanation! -- Ondrej Mosnacek Senior Software Engineer, Linux Security - SELinux kernel Red Hat, Inc.
