My context: I need to upgrade a production RKE2 cluster with pretty unreliable internet access. I would like to just have the cluster access an internal artifactory with the system images during upgrade without restricting (or at best temporarily) any running workloads.