Critical Description
in MR2 (Hadoop 0.23) the configuration used by a client to submit jobs to the cluster is used to configure the AM.
Currently this configuration is whatever has been set in hadoop-conf/ which is minimal (setting yarn framework).
A problem arises in how AM and HS communicate to each other info a about a job (via HDFS), AM using default conf values writes to /tmp while HS started by minicluster looks in target/test-dir thus producing random failures. Random because if the AM is still alive when the client checks for the job, the AM gives the info correctly, but if the AM is gone the HS does not know about the job.
We should seed the configuration of minicluster as the hadoop-conf/ used by the testcases.