[
https://issues.apache.org/jira/browse/HADOOP-5801?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robert Chansler updated HADOOP-5801:
------------------------------------
Description:
If the hosts file is changes across restart then it should be refreshed upon
recovery so that the excluded hosts are lost and the maps are re-executed.
The mapred-hosts list can change across restarts. Once the jobtracker recovers,
it detects all the tasktracker there are there in the history. If the hosts
list is changed, then the jobtracker will still have the tasktracker (data)
internally but will disallow the tracker when it contacts. As a result, the
jobtracker will have to wait for the tracker to timeout in order to re-execute
the tasks. This patch simply refreshes the node list upon recovery so that the
invalid trackers are lost immediately.
was:If the hosts file is changes across restart then it should be refreshed
upon recovery so that the excluded hosts are lost and the maps are re-executed.
Release Note: (was: The mapred-hosts list can change across restarts.
Once the jobtracker recovers, it detects all the tasktracker there are there in
the history. If the hosts list is changed, then the jobtracker will still have
the tasktracker (data) internally but will disallow the tracker when it
contacts. As a result, the jobtracker will have to wait for the tracker to
timeout in order to re-execute the tasks. This patch simply refreshes the node
list upon recovery so that the invalid trackers are lost immediately. )
Editorial pass over all release notes prior to publication of 0.21. Routine.
Bug.
> JobTracker should refresh the hosts list upon recovery
> ------------------------------------------------------
>
> Key: HADOOP-5801
> URL: https://issues.apache.org/jira/browse/HADOOP-5801
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Amar Kamat
> Assignee: Amar Kamat
> Fix For: 0.21.0
>
> Attachments: 5801-0.20.patch, HADOOP-5801-v1.2.patch
>
>
> If the hosts file is changes across restart then it should be refreshed upon
> recovery so that the excluded hosts are lost and the maps are re-executed.
> The mapred-hosts list can change across restarts. Once the jobtracker
> recovers, it detects all the tasktracker there are there in the history. If
> the hosts list is changed, then the jobtracker will still have the
> tasktracker (data) internally but will disallow the tracker when it contacts.
> As a result, the jobtracker will have to wait for the tracker to timeout in
> order to re-execute the tasks. This patch simply refreshes the node list upon
> recovery so that the invalid trackers are lost immediately.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
