Sorry I was wrong. I assumed based on Kubernetes' use of etcd it would be 3+ nodes. It turns out Kubernetes master is a single node currently which means they h

Sorry I was wrong. I assumed based on Kubernetes' use of etcd it would be 3+ nodes. It turns out Kubernetes master is a single node currently which means they haven't built high availability into the master at all... which is a pretty scary way to run a thing that manages your entire infrastructure. There's already a few topics on the mailing list about etcd losing its data and Kubernetes doesn't know how to recover. Yuck.