We moved to Centos 8 for user facing servers, from Fedora, during the pandemic. We needed more stability. We switched to AlmaLinux 8 a few months before the Centos 8 went EoL (using the AlmaLinux migration script). We now use AlmaLinux 9 and have no regrets.

On the topic of dnf vs. yum. From the admin PoV they're basically the same. Building RPMs works the same way, creating repos works the same way. I had to update our kickstart scripts. Most of our Puppet config was the same. The only problem was with older protocols being deprecated so we had to change the way we manage GPG keys. EL9 does not like older GPG keys, unless you switch to legacy crypto policies, but that only postpones the problem. We had keys on a per repository basis, with some repositories being shared across major versions. For EL9 we switch to having one key per major distribution version, with a separate set of repositories for each major version. We now use podman for building RPMs and creating repos, so it's not extra effort.

We also use Ubuntu, mostly because machine learning setups tend to be better supported on same. I've experimented with PXE autoinstall. It does the job but its nowhere near as powerful as kickstart. I have kickstart files that can generate a completely configured service VM just using the %post script. We only need Puppet for the ongoing management of systems. On Ubuntu I've not had much luck trying to get custom configuration to work in its YAML based config. I suspect we'll look at a push based system like Ansible for ongoing management on Ubuntu. We use Puppet because our EL9 clients are dual boot with Windows 11, and an agent works better in that setup.

I am grateful for Scientific Linux, but I'm not looking for further support for somebody else's distribution from its developers. There aren't the spare resources around any more to allow for that. They've got their own users and use cases to support now.