HyperShift currently runs 3 replicas of active/passive HA deployments such as kube-controller-manager, kube-scheduler, etc. In order to reduce the overhead of running a HyperShift control plane, we should be able to run these deployments with 2 replicas.

In a 3 zone environment with 2 replicas, we can still use a rolling update strategy, and set the maxSurge value to 1, as the new pod would schedule into the unoccupied zone.

    1.
    2.
    3.