hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Steve Loughran (JIRA)" <j...@apache.org>
Subject [jira] Created: (HADOOP-3880) TaskTracker doesnt recheck job tracker version on reconnect
Date Thu, 31 Jul 2008 14:34:34 GMT
TaskTracker doesnt recheck job tracker version on reconnect
-----------------------------------------------------------

                 Key: HADOOP-3880
                 URL: https://issues.apache.org/jira/browse/HADOOP-3880
             Project: Hadoop Core
          Issue Type: Bug
          Components: mapred
    Affects Versions: 0.19.0
            Reporter: Steve Loughran
            Priority: Minor


This isnt anything I have a test for/encountered, just something I noticed when reviewing
TaskTracker.
1. TaskTracker sets a justStarted flag to true when starting up
2. One way it uses this flag is to check job tracker versions -the version is only checked
when justStarted==true, which is reset after the check.
3. If a JobTracker is unreachable, then the TaskTracker sleeps for 5 seconds and then continues

There is a risk, therefore, that if the job tracker goes down, a different version might come
back up, and the Task Tracker will not notice until it makes an incompatible IPC call.

This is a pretty unlikely scenario; you've got to kill the job tracker and bring up a different
versioned one in 5 seconds. And the consequence of IPC incompatiblity will be a lost task,
regardless of whether this happens early or later in the process. I'm not sure it's worth
fixing. 

-- 
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