Details
-
New Feature
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
Description
In MAPREDUCE-824, we proposed introducing a hierarchy of queues in the capacity scheduler. Currently, the M/R framework provides the notion of job queues and handles some functionality related to queues in a scheduler-agnostic manner. This functionality includes:
- Managing the list of ACLs for queues
- Managing the run state of queues - running or stopped
- Displaying scheduling information about queues in the jobtracker web UI and job client CLI
- Displaying list of jobs in a queue in the jobtracker web UI and job client CLI
- Providing APIs for list queues and queue information in JobClient.
Since it would be beneficial to extend this functionality to hierarchical queues, this JIRA is proposing introducing the concept into the map/reduce framework as well. We could treat this as an umbrella JIRA and file additional tasks for each of the changes involved, sticking to the high level approach in this JIRA.
The motivation for hierarchical queues is discussed in the proposal on MAPREDUCE-824, for those interested.
Briefly, hierarchical queues allow administrators to have a greater control over how capacity (or conceivably other policies) associated with a queue can be used. They also allow delegation of control. Large clusters could have root level queues set up for organizations, and then operators from the organizations could be given access to manage queues under the root level queues (which we are calling sub-queues in
MAPREDUCE-824).