[HADOOP-5269] TaskTracker.runningTasks holding FAILED_UNCLEAN and KILLED_UNCLEAN taskStatuses forever in some cases.

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.19.1
• Fix Version/s: 0.19.2, 0.20.0
• Hadoop Flags: Reviewed
• Assignee: Amareshwari Sriramadasu
• Reporter: Amareshwari Sriramadasu
• Created: 17/Feb/09 05:35
• Updated: 08/Jul/09 16:53
• Resolved: 20/Feb/09 10:24

Description

Tasktracker is holdingup TaskStatus objects in runningTasks forever in somecases. This happens in the following scenario.
-> Task got an exception
-> Sets the phase to CLEANUP
-> The task tries to do cleanup. and it doesn't respond after that.
-> TaskTracker marks the task unresponsive and makes the task FAILED_UNCLEAN
-> TaskTracker doesn't remove it from runningTasks data structure, since phase is CLEANUP and state is FAILED_UNCLEAN (it treats this as cleanupAttempt).

I would propose that once the task goes to CLEANUP phase, kill on the task should mark it a clean failure i.e. The task state should be FAILED/KILLED.