The ingest framework support remote job scheduling capability current (through direct jobtracker interaction). This breaks when running MR on Yarn. (And yarn is where we want to go).

Options I can see are

• Drop support for MR1 (don't feel comfortable doing this quite yet, though can discuss)
• Maven munge/conditional compilation (might be the best option - I think it's tenable only because there's a sunset in sight (sunset being eventually dropping MR1 support))
• Moving over to Oozie for job scheduling. Adds another dependency on the cluster, but it's pretty minimal and common. I think this might be the right way to go. (I'm not stuck on Oozie particularly - that's just the only "answer" I'm aware of).

My vote is for the Oozie version (assuming I'm not missing something obvious)