[
https://issues.apache.org/jira/browse/MAPREDUCE-4302?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Daryn Sharp updated MAPREDUCE-4302:
-----------------------------------
Attachment: MAPREDUCE-4302.patch
Wrap log aggregation init with a try block. The log init is executed just
before localization starts instead of in parallel. Log init sends back an app
finish event on failure, or a log init success after which localization begins.
Log init is serialized because it only makes a few directories and spins off a
thread. If log init fails then the job itself is likely to fail, so a log init
failure reports a diagnostic of the exception back to the AM. Otherwise
if/when the job fails there will be no logs to debug...
> NM goes down if error encountered during log aggregation
> --------------------------------------------------------
>
> Key: MAPREDUCE-4302
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-4302
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 0.23.0, 2.0.0-alpha, trunk
> Reporter: Daryn Sharp
> Assignee: Daryn Sharp
> Priority: Critical
> Attachments: MAPREDUCE-4302.patch
>
>
> When a container launch request is sent to the NM, if _any_ exception occurs
> during the init of log aggregation then the NM goes down. The problem can be
> induced by situations including, but certainly not limited to: transient rpc
> connection issues, missing tokens, expired tokens, permissions, full/quota
> exceeded dfs, etc. The problem may occur with and without security enabled.
> The ramification is an entire cluster can be rather easily brought down
> either maliciously, accidentally, or via a submission bug.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
