Requested by hagleitn.
Essentially, users of TezClient (Hive in this case) would create a Tez session. This session can then be used to submit multiple DAGs.
Internally, TezClient can control whether an AM is run within the same process, in unmanaged mode, on the cluster etc. Initially, we'd likely end up creating one AM per submitted DAG. Once
TEZ-340 is in place, the same AM could be re-used for multiple non-concurrent DAGs.