It’s intentional that we only support one control URL. This is not currently implemented, but the proxy is intended to resolve the control url’s authority to a set of IP addresses and load balance requests over all available nodes (as it does when it communicates with any other Kubernetes service).
Currently, when the proxy fails to connect to the control url, it will re-resolve the hostname to a new IP address; so the proxy generally connects quickly to new instances (i.e. when replacing the controller pod). Furthermore, in Kubernetes, the destination service is served behind a ClusterIP (so kubeproxy load balancing applies)