I believe using a headless service on Kubernetes is the right way to build node discovery and auto-clustering on that system, but I'm interested in learning about other approaches.