[ https://issues.apache.org/jira/browse/MAPREDUCE-969?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon updated MAPREDUCE-969:
----------------------------------
Attachment: reduce_task_logs
bad_job_jt_logs
bad_job_events
Attaching sanitized logs from this incident.
The event that seems to be a red flag is the lost task tracker xx05.
The null pointer exception is caused by u.getHost() being null - this URI is the taskTrackerHttpAddress
in a TaskTrackerStatus. The job event output doesn't show any with a malformed URL, so I suspect
some kind of race.
Aside from this issue, I find it odd that GetMapEventsThread ignores exceptions. In cases
like this it will cause the ReduceTask to spin forever while still reporting progress until
the user intervenes.
> NullPointerException during reduce freezes job
> ----------------------------------------------
>
> Key: MAPREDUCE-969
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-969
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: jobtracker, task, tasktracker
> Affects Versions: 0.20.2
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Attachments: bad_job_events, bad_job_jt_logs, reduce_task_logs
>
>
> We experienced several jobs stuck in Reduce on a cluster. All of the stuck reduce tasks
had a similar were stuck at "Need another 2 map output(s) where 0 is already in progress"
despite all of the mappers having completed, and 0 scheduled. The stuck reducers had experienced
the following exception early in the shuffle:
> java.lang.NullPointerException
> at java.util.concurrent.ConcurrentHashMap.get(ConcurrentHashMap.java:768)
> at org.apache.hadoop.mapred.ReduceTask$ReduceCopier$GetMapEventsThread.getMapCompletionEvents(ReduceTask.java:2747)
> at org.apache.hadoop.mapred.ReduceTask$ReduceCopier$GetMapEventsThread.run(ReduceTask.java:2670)
> Will attach more information and logs momentarily.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
|