[
https://issues.apache.org/jira/browse/HADOOP-11138?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
shreyas subramanya updated HADOOP-11138:
----------------------------------------
Attachment: HADOOP-11138-trunk.patch
Hi Allen,
I have modified the patch for the trunk. There are 2 parts to it:
1. For streaming container logs, we set the following in mapred-site.xml:
"yarn.container.log.streaming.enabled",
"yarn.container.log.streaming.port"
2. For daemon logs streaming (NN, DN, journal, zkfc, RM, NM, proxyserver,
timelineserver, historyserver), we set the socket appender for the root logger
in hadoop-env.sh:
export HADOOP_DAEMON_ROOT_LOGGER=INFO,RFA,SOCKA
If step 2 were to be done purely using properties, all the daemon classes need
to be modified to add the socket appender to root logger programmatically.
Since the current approach is not adding any more environment variables (unless
the user wants to override the ports), I think this is a better approach that
modifying the daemon classes directly.
> Stream yarn daemon and container logs through log4j
> ---------------------------------------------------
>
> Key: HADOOP-11138
> URL: https://issues.apache.org/jira/browse/HADOOP-11138
> Project: Hadoop Common
> Issue Type: Improvement
> Components: util
> Affects Versions: 2.6.0
> Reporter: shreyas subramanya
> Priority: Minor
> Attachments: HADOOP-11138-1.patch, HADOOP-11138-trunk.patch,
> HADOOP-11138.patch
>
>
> Resource manager, node manager, history server, application master and other
> container syslogs can be streamed by configuring a log4j SocketAppender in
> the root loggers.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
