[jira] Commented: (HADOOP-153) skip records that fail Task

["Doug Cutting (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-153?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12616554#action_12616554 ] 

Doug Cutting commented on HADOOP-153:
-------------------------------------

> Since we are handling task crashes, the tasktracker and jobtracker have to get involved at some level, no?

You're right, at least the TaskTracker needs to be involved, if it's to capture crashed processes, and the JobTracker if it's to handle machine crashes.  Sigh.

A few issues with the patch:
 - InterTrackerProtocol's version needs to change since the format of a Task and TaskStatus have changed.
 - UmbilicalProtocol#reportNextRecordRange() needs javadoc.  This javadoc is not published, but our protocols should be well-documented.
 - You add a method to the Reporter interface: are we sure no users implement this?  Also, this is the wrong place for this method.  It has nothing to do with reporting status.  Unfortunately, getInputSplit() was added to Reporter at some point in the past, setting a bad example.  These should be added to the context (once we have contexts).
 - in MapRunner, autoIncrPrcossedRecord is a typo.



> skip records that fail Task
> ---------------------------
>
>                 Key: HADOOP-153
>                 URL: https://issues.apache.org/jira/browse/HADOOP-153
>             Project: Hadoop Core
>          Issue Type: New Feature
>          Components: mapred
>    Affects Versions: 0.2.0
>            Reporter: Doug Cutting
>            Assignee: Sharad Agarwal
>         Attachments: 153_1.patch, 153_2.patch, skipRecords_wip1.patch
>
>
> MapReduce should skip records that throw exceptions.
> If the exception is thrown under RecordReader.next() then RecordReader implementations should automatically skip to the start of a subsequent record.
> Exceptions in map and reduce implementations can simply be logged, unless they happen under RecordWriter.write().  Cancelling partial output could be hard.  So such output errors will still result in task failure.
> This behaviour should be optional, but enabled by default.  A count of errors per task and job should be maintained and displayed in the web ui.  Perhaps if some percentage of records (>50%?) result in exceptions then the task should fail.  This would stop jobs early that are misconfigured or have buggy code.
> Thoughts?

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.