An administrator is designing a kubeadm cluster with several control-plane nodes. How should clients reach a control plane that remains accessible when one API server fails?

1. Use a stable controlPlaneEndpoint backed by a load balancer for the API servers
2. Set the endpoint to the first worker's Pod CIDR address
3. Expose each kubelet as an interchangeable API server
4. Use a ClusterIP Service as the only bootstrap endpoint before any control plane exists