[
https://issues.apache.org/jira/browse/HADOOP-5201?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12672002#action_12672002
]
Steve Loughran commented on HADOOP-5201:
----------------------------------------
Is there a currently a requirement that the application reading the task report
has 100% of all exceptions that may get thrown on the server? If not, you end
up with problems at the far end trying to unmarshall something that you can't
handle.
This is a recurrent problem in RMI and SOAP stacks, where you get some error
message telling you that something went wrong at the far end but they can't
tell you what it was. It may be better to serve up some portable summary of the
exception (classname, message, stack) without trying to push the full exception
over any wire
> getDiagnostics in TaskReport should return exceptions
> -----------------------------------------------------
>
> Key: HADOOP-5201
> URL: https://issues.apache.org/jira/browse/HADOOP-5201
> Project: Hadoop Core
> Issue Type: Improvement
> Components: mapred
> Reporter: Santhosh Srinivasan
>
> Currently, getDiagnostics() returns Strings. When exceptions are thrown in
> user code and/or Hadoop, it would be cleaner to propagate the exception back
> to the application for better error handling. Hadoop should return the
> exceptions instead of returning string representations that correspond to
> printStackTrace() output.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
