fleetctl usability issue from worker nodes

[sigma]

Given a cluster in the following configuration: https://coreos.com/docs/cluster-management/setup/cluster-architectures/#production-cluster-with-central-services

Attempting to run "fleetctl list-machines" from a worker node gives an empty list.
The reason appears to be that while the fleet service is properly pointing to external etcd nodes, fleetctl still has a default endpoint that points to localhost:4001 for etcd (which actually should not even exist in this configuration)

I think it would be nice if instead of hardcoding the default etcd endpoint, etcdctl was smart enough to look for a locally running fleetd and try to get the etcd node(s) from there (maybe via /var/run/fleet.sock ?). After all, running fleetctl from a node of a cluster should probably default to talking to that cluster, regardless of where etcd is.

[bcwaldon]

@sigma We are in the process of migrating to the fleet API. Once we merge #1034, the fleetctl tool will no longer access etcd directly. It will use the local /var/run/fleet.sock.