What are the most critical components of a Kubernetes cluster when troubleshooting node failures at P2P?