Take snapshots of the cluster etcd and store them off-node
Currently the entire state of the cluster, as stored in etcd, does not exist anywhere except on the disk on the master node. If that node crashes or the disk gets corrupted, there is no recovery mechanism.
A rudimentary recovery mechanism should be added to backup the etcd datastore periodically and store those backups off the node.