The CPO does not currently respect the CVO runlevels as standalone OCP does.

The CPO reconciles everything all at once during upgrades which is resulting in FeatureSet aware components trying to start because the FeatureSet status is set for that version, leading to pod restarts.

It should roll things out in the following order for both initial install and upgrade, waiting between stages until rollout is complete:

• etcd
• kas
• kcm and ks
• everything else