Infrastructure happenings, second half of aug - first half of sept 2024
So, I was going to try and do these posts more regularly, but of course thats hard to do. After flock there was a bunch of things I wanted to post, then a bunch of fires and so things got behind. Such is life, so here's a few things I wanted to talk about in more detail from the last month or so. As always, I do still post on mastodon daily, happy to answer questions or comments there as things happen and expand on things in posts like this.
Fedora 41 branched off rawhide! This I think went much more smoothly than the last cycle. I like to hope it's because we documented all the things that were not right last time and did them this time. There were a few more things to adjust, it wasn't perfect, but it was much better!
We upgraded our OpenShift cluters from 4.15 to 4.16. I continue to be very happy how smooth OpenShift upgrades are. Not 100% seamless, but pretty good. This time we had some storage stuff that caused the upgrade to not finish, but it wasn't too hard to work around. So much nicer than the old 3.x days.
We landed a bunch of koji/kiwi changes before Beta freeze. Kudos to Neal Gompa and Adam Williamson for working through all those. It was nice to mostly get everything lined up before Freeze so we didn't have to be doing a lot of churn. We got everything working in rawhide first, then merged the f41 changes.
Had a really anoying IPA outage. I was running our main playbook (runs over everything) on a thursday night, just to make sure everything was in sync for the freeze, and... our playbook thought all our ipa servers were not configured right and tried to uninstall and resync them all. Luckily the server that was the CA master refused to uninstall, so we were still up on one server. From that we were able to reinstall/resync the other 2 and get things back up and working. I am still not sure why the playbook saw no dirserv running on the servers (and thus thought they were unconfigured). We are going to adjust that playbook to definitely not try and do that, and instead move setting up a replica to a manual playbook only run by humans as needed.
Thanks to a bunch of work from Stephen Gallagher and Carl George, eln and epel10 are now doing composes just like we do for rawhide and branched. This should allow us to retire our old ODCS (on demand compose service) setup, as its not really maintained upstream anymore and is on EOL os versions. Great to get things all running the same way, but of course we will probibly change everything next year or something.
We managed to sign off on Fedora 41 Beta being released next week. I was pretty amazed, as it didn't seem like we had enough time to really shake out all the bugs, but testing coverage ended up being pretty good. Looking forward to Beta next week and end of Beta freeze.