If upgrading a node to RHEL9 via osImageURL override that is otherwise RHEL8 (e.g. OCP 4.12 RHCOS) then because rhel9's systemd enabled new features in the journal file, on rollback to rhel8 systemd will refuse to read the journal file.

This breaks the way the MCO does state management by writing to the journal (pending config, etc.)

I think the main fix for this is to go back to having an out of band file.

(Ultimately of course, I think the node state should actually be maintained by e.g. bootc, which itself would be responsible for transactional updates of OS and config)

4.13

Always

1. Launch 4.12 cluster, use osImageURL override for latest digest in quay.io/openshift-release-dev/ocp-v4.0-art-dev:4.13-9.2 2.
2. Wait for MC rollout
3. Remove override
4. Observe degraded node