[MAPREDUCE-7198] mapreduce.task.timeout=0 configuration used to disable timeout doesn't work

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 3.3.0
• Target Version/s: 3.3.0
• Assignee: Akira Ajisaka
• Reporter: Akira Ajisaka
• Created: 23/Apr/19 02:34
• Updated: 23/May/19 01:37
• Resolved: 23/May/19 01:22

Description

mapreduce.task.timeout=0 configuration used to disable timeout doesn't work after MAPREDUCE-6190. If the task timeout is configured as zero the task fails with stuck timeout, if the TaskStatus is null.

          if (sendProgress) {
            // we need to send progress update
            updateCounters();
            checkTaskLimits();
            taskStatus.statusUpdate(taskProgress.get(),
                                    taskProgress.toString(),
                                    counters);
            amFeedback = umbilical.statusUpdate(taskId, taskStatus);
            taskFound = amFeedback.getTaskFound();
            taskStatus.clearStatus();
          }
          else {
            // send ping 
            amFeedback = umbilical.statusUpdate(taskId, null);
            taskFound = amFeedback.getTaskFound();
          }

This issue is reported by bibinchundatt.

Issue Links

• is broken by: MAPREDUCE-6190 If a task stucks before its first heartbeat, it never timeouts and the MR job becomes stuck