mesos-dns fails to start if any ZK server in the configured ensemble is not running

[jdef]

I killed my second ZK in a 3 member cluster then tried starting mesos-dns and it failed to boot:

VERY VERBOSE: 2017/07/06 22:00:32 resolver.go:670: Starting master detector for ZK  zk://10.2.0.5:2181,10.2.0.6:2181,10.2.0.7:2181/mesos
VERY VERBOSE: 2017/07/06 22:00:32 resolver.go:150: Warning: waiting for initial information from Zookeper.
vagrant@node-1:~/mesos-dns-cleanroom/src/github.com/mesosphere/mesos-dns$ 2017/07/06 22:00:33 Failed to connect to 10.2.0.6:2181: dial tcp 10.2.0.6:2181: i/o timeout

mesos-dns shouldn't die in this case because 2/3 of the ZK cluster is running.

perhaps the default ZK timeout is too short?