Comment 1 for bug 1543216

It looks like this is the same underlying cause as bug #1538583. Both the manual provider and maas provider change the mongo replicaset after bootstrap completes, causing the apiserver to drop all its connections. Here's how it happened here:

Client logs in fine:
2016-02-08 15:12:47 DEBUG juju.apiserver apiserver.go:287 -> [11] user-admin@local 57.72784ms {"RequestId":1,"Response":"'body redacted'"} Admin[""].Login

Client does a ModelGet just fine (part of the deploy command)
2016-02-08 15:12:47 DEBUG juju.apiserver apiserver.go:273 <- [11] user-admin@local {"RequestId":3,"Type":"Client","Version":1,"Request":"ModelGet","Params":"'params redacted'"}
2016-02-08 15:12:47 DEBUG juju.apiserver apiserver.go:287 -> [11] user-admin@local 3.237235ms {"RequestId":3,"Response":"'body redacted'"} Client[""].ModelGet

But then we see the kiss of death for all api connections:
2016-02-08 15:12:48 INFO juju.apiserver apiserver.go:504 got error pinging mongo: EOF

After this point the apiserver will die and kill all its connections, so the next time the deploy command tries to use the API connection it has, it will get an error that it is using a closed connection ("connection is shut down").