hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Rick Cox (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HADOOP-2057) streaming should optionally treat a non-zero exit status of a child process as a failed task
Date Mon, 15 Oct 2007 17:08:50 GMT

     [ https://issues.apache.org/jira/browse/HADOOP-2057?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Rick Cox updated HADOOP-2057:
-----------------------------

    Attachment: exit-status-2057.patch

If the premise of this change is acceptable, than I'd like to request a code-review of this
patch. 

It:
* supports the stream.non.zero.exit.is.failure job configuration setting in PipeMapRed
* adds a mention about that setting to the -info text in StreamJob
* adds a test case
* adds stream.non.zero.exit.is.failure to hadoop-default.xml, with a backwards-compatible
default of false


> streaming should optionally treat a non-zero exit status of a child process as a failed
task
> --------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-2057
>                 URL: https://issues.apache.org/jira/browse/HADOOP-2057
>             Project: Hadoop
>          Issue Type: Improvement
>          Components: contrib/streaming
>    Affects Versions: 0.14.2
>            Reporter: Rick Cox
>         Attachments: exit-status-2057.patch
>
>
> The exit status of the external processes spawned by streaming tasks is currently logged,
but not used to indicate success or failure of the task. While this is reasonable for some
UNIX tools (e.g. grep), many programs will indicate failure by a non-zero exit status. (Also,
even for custom programs, intentionally indicating the failure of a streaming task is currently
rather tricky.)
> This could be supported by adding a new job-configuration setting, 'stream.non.zero.exit.is.failure'.
If true, a non-zero exit status of a child process would throw an exception in the PipeMapRed,
causing task failure. The current behavior would be preserved by using a default setting of
false. 
> This would allow streaming tasks to easily indicate failure, even if all input has already
been consumed.

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


Mime
View raw message