[
https://issues.apache.org/jira/browse/HADOOP-1605?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12529266
]
Koji Noguchi commented on HADOOP-1605:
--------------------------------------
I'm probably missing something, but I don't see much benefit of running a
watcher at the Java level. Maybe a shell script? (which is probably what the
existing tools already do)
With your patch, hadoop-daemon.sh would save the pid of the watcher process.
But to guarantee that we don't bring up multiple namenodes, doesn't it need to
check both the watcher pid and namednode pid?
1) If I stupidly kill -9 the watcher
2) If namenode somehow ignores the process.destroy() which is probably a
SIGTERM in linux.
> Automatic namenode restart when it encounters an error situation
> ----------------------------------------------------------------
>
> Key: HADOOP-1605
> URL: https://issues.apache.org/jira/browse/HADOOP-1605
> Project: Hadoop
> Issue Type: Improvement
> Components: dfs
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Fix For: 0.15.0
>
> Attachments: namenodeRestart4.patch, watcher.patch
>
>
> The namenode dies when it encounters an unexpected Runtime Exception.
> Instead, it can catch exceptions, clears up all its internal data structures
> and restarts. This was attempted in HADOOP-1486 earlier.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
