This message was deleted.

It depends on the upgrade. If there are changes/upgrades to the CNI chart that can be disruptive. In general I wouldn’t expect an upgrade to a single-node cluster to be entirely non-disruptive, if you want that you should go for at least 3 servers in a HA configuration.

So this “cluster” uses Calico for its CNI; the calico-node and other pods haven’t been restarted since March. It seems only kubelet, kube-proxy, and the control plane components were changed in this case. I agree, though, that ideally we’d have a cluster of at least 3 nodes for true HA - I was just hoping to avoid unnecessary issues during routine updates.