We'd really like not to "nuke" the cluster, so any help about this failed upgrade, which lead to a failing cluster, is highly appreciated.