Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Cannot Reproduce
-
2.2.0
-
None
-
None
Description
While running Spark streaming applications on YARN in cluster mode, reboot/shutdown of the node hosting AM causes the application to terminate SparkContext and mark it as SUCCEEDED. This is misleading as:
- streaming applications are long-running and not expected to terminate
- setting spark.yarn.maxAppAttempts for application re-attempt does not take effect since this scenario is not recognized as a failure
Please provide a workaround for resubmitting streaming/long-running applications in the case of AM shutdown.