Mailing-List: contact core-dev-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: core-dev@hadoop.apache.org
Message-ID: <562078869.1206492444907.JavaMail.jira@brutus>
Date: Tue, 25 Mar 2008 17:47:24 -0700 (PDT)
From: "Rick Cox (JIRA)" <jira@apache.org>
To: core-dev@hadoop.apache.org
Subject: [jira] Updated: (HADOOP-3089) streaming should accept stderr from
 task before first key arrives
In-Reply-To: <604990250.1206456565827.JavaMail.jira@brutus>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


     [ https://issues.apache.org/jira/browse/HADOOP-3089?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rick Cox updated HADOOP-3089:
-----------------------------

    Attachment: patch-stderr-3089.txt

Patch starts {{MRErrorThread}} immediately, adds a reporter later.

{{testStderrNoInput()}} fails without this patch, and passes with it on Linux (RHEL3) and Mac OS X. It does make some assumptions about how large a stderr buffer might be, but that's unfortunately an innate component of this issue.

{{testStderrCountsAsProgress()}} passes, but isn't meaninful because {{LocalJobRunner}} does not seem to support {{mapred.task.timeout}}.

> streaming should accept stderr from task before first key arrives
> -----------------------------------------------------------------
>
>                 Key: HADOOP-3089
>                 URL: https://issues.apache.org/jira/browse/HADOOP-3089
>             Project: Hadoop Core
>          Issue Type: Bug
>          Components: contrib/streaming
>    Affects Versions: 0.14.0, 0.14.1, 0.14.2, 0.14.3, 0.14.4, 0.15.0, 0.15.1, 0.15.2, 0.15.3, 0.16.0, 0.16.1
>            Reporter: Rick Cox
>            Assignee: Rick Cox
>             Fix For: 0.17.0
>
>         Attachments: patch-stderr-3089.txt
>
>
> Stderr output from a streaming task is not collected until the {{MRErrorThread}} is started by {{PipeMapRed.startOutputThreads()}}, which is done on the first call to {{map()}} or {{reduce()}}.
> This makes it difficult to debug failures in starting up the task process. It can also lead to deadlock when a task receives no input keys but produces significant stderr output: the process will block on writing to stderr, while streaming will block waiting for the process to exit.
> We should start the {{MRErrorThread}} when the process is forked, and then add the {{reporter}} later to enable stderr output serve as a keep-alive.

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