[ http://issues.apache.org/jira/browse/HADOOP-303?page=all ]
Benjamin Reed updated HADOOP-303:
---------------------------------
Attachment: jobcl-fix.patch
> JobClient looking for classes for submitted job in the wrong place
> ------------------------------------------------------------------
>
> Key: HADOOP-303
> URL: http://issues.apache.org/jira/browse/HADOOP-303
> Project: Hadoop
> Type: Bug
> Components: mapred
> Versions: 0.3.2
> Reporter: Benjamin Reed
> Priority: Minor
> Attachments: jobcl-fix.patch
>
> JobClient does some checking of the job being submitted when it submits a jar
> file along with the job. The problem is that the JobClient pulls classes from
> the classpath rather than the submitted jar file. Because the jar file may
> contain newer (or older) versions of classes on the classpath this behavior
> leads to confusing errors when the job is run. It is also a pain to ensure
> that the jar file being submitted is on the classpath. Further, if the
> JobClient uses the submitted jar file rather than the classpath, missing
> classes from the jar file can be detected earlier.
> This patch will cause the JobClient to load the classes for a job from the
> jar file rather than the classpath. Because of the class loading precedence
> rules in Java, if the class is on the system class path, it will be loaded
> from there rather than the submitted jar file, but now users need not (and
> should not) put job classfiles on the system classpath.
> This patch also allows config files to be put in a configuration directory
> rather than on the classpath, which also eliminates some confusing behavior
> when there are duplicate instances of config files in different parts of the
> classpath.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
