There isn't a definitive way to do this. However, as long as the etcd data and the rest of the state of the cluster is stored somewhere, then this shouldn't be impossible. Upgrading nodes wouldn't be too much of a concern due to the fact that the pods will run on another node during the switch. But the master is where I'd worry. You should first upgrade the master by spinning up a new one with the newer kubernetes, then replicating state of the old master. I'm a little fuzzy on backwards capabilities with the coming 1.1 but hopefully you can then connect each node to the new master and upgrade those one by one with little to no issues.