[
https://issues.apache.org/jira/browse/HADOOP-5136?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12668013#action_12668013
]
Steve Loughran commented on HADOOP-5136:
----------------------------------------
Likely cause is
# {{localJobConf}} is set in {{TaskInProgress.launchTaskForJob()}}, which
happens at the end of {{TaskInProgress.localizeJob()}}; if localizeJob() fails
early due to some problem (in my case, HDFS not being fully live), then the
class is not fully initialized.
If the TT gives up on the task and calls cleanup(true), then the task NPEs.
> NPE in TaskInProgress.cleanup
> -----------------------------
>
> Key: HADOOP-5136
> URL: https://issues.apache.org/jira/browse/HADOOP-5136
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.21.0
> Reporter: Steve Loughran
> Priority: Minor
>
> This may be something that only my code triggers; an NPE in
> TaskTracker$TaskInProgress.cleanup
> {code}
> [sf-startdaemon-debug] 09/01/28 11:41:06 [TaskLauncher for task] INFO
> mapred.TaskTracker : Error cleaning up task runner:
> java.lang.NullPointerException
> [sf-startdaemon-debug] at
> org.apache.hadoop.mapred.TaskTracker$TaskInProgress.cleanup(TaskTracker.java:2487)
> [sf-startdaemon-debug] at
> org.apache.hadoop.mapred.TaskTracker.startNewTask(TaskTracker.java:1825)
> [sf-startdaemon-debug] at
> org.apache.hadoop.mapred.TaskTracker.access$1100(TaskTracker.java:104)
> [sf-startdaemon-debug] at
> org.apache.hadoop.mapred.TaskTracker$TaskLauncher.run(TaskTracker.java:1779)
> {code}
> Looking at the code, the only source of NPE's on that line is localJobConf
> {code}
> if (localJobConf.getNumTasksToExecutePerJvm() == 1) {
> {code}
> It looks like if TaskInProgress.cleanup() ever gets called with no valid
> localJobConf, then an NPE is the result. The exception gets logged and
> discarded, but it does appear in the logs.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
