Using kubectl scale to scale control plane nodes

[marklewin]

I've been able to scale nodes up/down using the kubectl scale command. However, when scaling down the number of CP replicas, I noticed:

• It throws an error if the operation would result in an odd number of replicas
• If I scale back to a single replica, it borks the cluster

For comparison, I used kubectl edit to scale up and down the number of worker nodes and also control plane nodes. I was able to go from 3 CP nodes to 1 CP node, seemingly without error. It didn't let me edit the config to have two CP nodes though.

kubectl-scale.txt
kubectl-edit-workers.txt
kubectl-edit-cp.txt

[George-Aeillo]

It is expected that it doesn't let you have two control plane nodes. This would cause a "split brain" issue, where etcd would not be able to determine a leader. When finding this problem, did you use the OCI provider on OCNE 2.0? Were you running a managed or self-managed setup for your cluster?