Hdfs framework failure on restarting cluster

[mezosfer]

I have an 8 node cluster on virtualbox running 3 JN, 2 NN, 2ZK and 2 DN. I deployed HDFS and Spark on top of them. Everything works fine. After i restarted the machines mesos, marathon and chronos frameworks are ok, but when i tried to re-run hdfs-mesos, i've encountered some problems. It couldn't launch name nodes and data nodes. It launced 3 JN and 6 ZK instead of them. Here is the stderr of namenode failures:

WARN hdfs.DFSUtil: Exception in creating socket address :50071
java.lang.IllegalArgumentException: Does not contain a valid host:port authority: :50071

So, what's the proper way of running hdfs-mesos after restarting the host machines?