Oguz,
Depends.
The simplest way would be to place an agent on one or more of the Hadoop servers.
If Hadoop automatically distributes load across nodes, this is likely to be sufficient.
If Hadoop can be invoked via a web service call, then no need for a Hadoop 'local' OS agent. You will need the UC4 WebService agent on some device.
You will need to test exactly how Hadoop presents status to UC4.
One that provides limited capabilities and Automic will not suggest - if the Hadoop server(s) will allow remote script submission at the OS level.... You will need to test this to see EXACTLY how the submission is done and if the connection is maintained until the submission is complete OR until the launched process is complete