Description
After launching an application through yarn-cluster mode, the SparkSubmit process sticks around and enters a monitoring loop to track the application's status. This is really a responsibility that belongs to a different process, such that SparkSubmit can run yarn-cluster applications in a fire-and-forget mode. We currently already do this for standalone-cluster mode. we should do it for yarn-cluster mode too.
Attachments
Issue Links
- is related to
-
SPARK-6692 Add an option for client to kill AM when it is killed
- Closed
- links to