Right now, there's a one-to-one association between a Kubernetes cluster and etcd. That's limiting in what concerns high-availability.

I for one, would like to have my etcd cluster outside of Kubernetes nodes, living for the sole purpose of providing the service to whatever apps I want, like a couple Kubernetes clusters (one for project A, another for project X).

This is not possible because etcd is not namespaced by any sort of cluster ID, thus clusters A & B data would overlap in a destructive way.

What I propose is to:

• Namespace etcd by some sort of unique cluster ID
• Implement apiserver support for master-slave operation, thus guaranteeing apiserver high-availability