If you are a epel6 or epel7 user on the ppc64 platform, I have some sad news for you. If you aren’t feel free to read on for a tale of eol architectures.

ppc64 (the big endian version of power) was shipped with RHEL6 and RHEL7 and Fedora until Fedora 28. It’s been replaced by the ppc64le (little endian) version in Fedora and RHEL8.

The fedora build system (koji) runs Fedora on all it’s builders. We moved to doing this because we often needed new features in rpm or lower level packages like that. For example when rpm switched to xz compression, we needed a rpm package that was new enough to do/understand that on all the builders, so we either had to backport this support to the RHEL version or just switch to Fedora. We found it more supportable to just switch to Fedora.

However, since Fedora stopped supporting ppc64 in Fedora 29, all our current ppc64 builders are Fedora 28 (the last release with support for ppc64). Fedora 28 is about to go end of life and we had to decide what to do about epel6/epel7 since they still support ppc64.

epel6 ppc64 users may be sticking with that platform because their hardware doesnt support ppc64le. epel7 users could move to ppc64le, but they might be keeping ppc64 instances to remain compatible or not wanting to change. ppc64 users are a very low number (around 100 checkins to our mirroring system per day, next to 1.5million for x86_64). Additionally, I would expect few of those ppc64 installs are new deployments or even directly on the internet.

We could make RHEL7 builders just for epel6/7 ppc64. However, our ansible playbooks for builder deployment have counted on them being Fedora for a long time now, so it would be a pretty big retooling effort. Additionally, those builders would sit around mostly idle, taking up resources we could use for other ones.

So, in the end, I think we are going to look at retiring ppc64 in epel6/7 next week when Fedora 28 goes end of life. The old packages would still be available in koji, but the repos would disappear. If there’s some case we haven’t thought of here, do bring it up to the epel steering comittee: https://pagure.io/epel/issue/57 or on the epel-devel list.