[TEZ-676] Tez job fails on client side if nodemanager running AM is lost

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.4.0
• Assignee: Hitesh Shah
• Reporter: Yesha Vora
• Created: 11/Dec/13 08:20
• Updated: 30/Mar/14 08:01
• Resolved: 10/Mar/14 23:17

Description

Scenario:

1) Run a long running Teragen Job
2) Find out the node where AM has started.
3) Kill nodemanager on AM host using kill -9 command

Expected:
2nd AM should be started and Job should be resumed. Job should also keep running on client side

Actual:
Here, the 1st am was started and then NM running AM was killed. The job wait for around 10 min to start 2nd AM. After that, 2nd AM attempt was started. Just at the same time, job output says that "job failed" and it exited.
Though RM has already started 2nd AM. Gradually 2nd AM runs are job finishes successfully.

Issue Links

• duplicates: TEZ-36 TaskScheduler should not unregister from the RM when sent a reboot
• is related to: TEZ-868 AM unregisters when sent a SIGTERM even though it is not the last attempt
• is related to: TEZ-36 TaskScheduler should not unregister from the RM when sent a reboot
• is related to: TEZ-289 DAGAppMasterShutdownHook should not report KILLED when exception brings down AM