Race condition in re-init of TaskTracker

• Key: HADOOP-4924
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Blocker
• Assignee: Devaraj Das
• Reporter: Devaraj Das
• Fix Version/s: 0.18.3, 0.19.1

Description

The taskReportServer is stopped in the TaskTracker.close() method in a thread. The race condition is:
1) TaskTracker.close() is invoked - this starts a thread to stop the taskReportServer
2) TaskTracker.initialize is invoked - this tries to create a new taskReportServer
Assume that the thread started to stop the taskReportServer gets to start its work after (2) above. The thread will end up stopping the newly created taskReportServer.

Comments

[Devaraj Das]

In the attached patch, I removed the thread doing the taskReportServer.stop(). Instead the TaskTracker.close() stops the taskReportServer inline (this is anyway required for the case where the config for the TaskTracker specifies the RPC port as some non-zero number).

[Arun C Murthy]

+1

[Devaraj Das]

All tests including test-patch passed on my machine. Committed this.

[dhruba borthakur]

It would be nice if somebody can comment if this affects 0.17 and 0.18 as well.

[Devaraj Das]

Yes it does affect those versions as well.

[dhruba borthakur]

Hi Devaraj, If this affects 0.19 as well, does it need to be commtted into that branch as well?

[Devaraj Das]

Ok I committed this to 0.19 branch too..

[dhruba borthakur]

Thanks Devaraj, for committing this into 0.19.

[Devaraj Das]

I committed this to the 0.18 branch too.