[proposal] Koordinator supports working with Apache Hadoop YARN

[zwzhang0107]

What is your proposal:
Koordinator supports working with YARN for resource sharing, so that batch workloads in YARN can share the unused resources of K8s workloads.

Why is this needed:
Although K8s has provided well-designed framework for batch workloads, there are still a lot of BigData jobs running on YARN in reality for some reasons. In order to improve cluster resource utilization, we need to support Koordinator working with YARN for hybrid workloads orchestration.

Is there a suggested solution, if so, please add it:

[FillZpp]

/assign @FillZpp @zwzhang0107

This is a general issue and it should be splitted into multiple sub-task issues to track after we have got an agreement on the approach.