The kubelet restart at the end of the join script appears
to be unnecessary, since the only action taken by the script
between that and the previous kubelet start is node labelling
(which doesn't require a kubelet restart).
In addition, the timing of this restart may be triggering
a kubernetes state synchronization bug, where a pod's status
isn't updated to reflect the readiness of all of its containers.
Change-Id: I480d1b345e5ddcce0cac961ff9c2b76526c5b76f
153882b153