When running an “external” etcd cluster from Kubernetes every Kubernetes upgrade needs to be grouped with an etcd upgrade as well (if required). In other words, etcd and k8s are not on separate upgrade paths in these cases. This might now be obvious to people who are used to running etcd in a pod within Kubernetes as is the default when setting up with kubeadm. It also means that the entire Kubernetes cluster must be taken down, or that new etcd nodes need to be independently setup before the Kubernetes upgrade can be pointed to them.