[
https://issues.apache.org/jira/browse/YARN-3269?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15994501#comment-15994501
]
stefanlee commented on YARN-3269:
---------------------------------
thanks for this jira, but when we do not config
*yarn.nodemanager.remote-app-log-dir*, it will be */tmp/logs* and NM or mr
historyserver will check */tmp/logs* 's scheme when write log to HDFS or
read log from HDFS,then they will throw exception of {{No AbstractFileSystem
for scheme: null}} in class *AbstractFileSystem.java* ,so i suggest that when
*yarn.nodemanager.remote-app-log-dir* is default value(scheme is null ), we
still use *FileContext.getFileContext(conf)* in *AggregatedLogFormat.java*,
else we can use *FileContext.getFileContext(remoteAppLogFile.toUri(), conf)* ,
am i wrong with this question? [~xgong] [~zhz]
> Yarn.nodemanager.remote-app-log-dir could not be configured to fully
> qualified path
> -----------------------------------------------------------------------------------
>
> Key: YARN-3269
> URL: https://issues.apache.org/jira/browse/YARN-3269
> Project: Hadoop YARN
> Issue Type: Bug
> Reporter: Xuan Gong
> Assignee: Xuan Gong
> Fix For: 2.8.0, 2.7.4, 3.0.0-alpha1
>
> Attachments: YARN-3269.1.patch, YARN-3269.2.patch
>
>
> Log aggregation currently is always relative to the default file system, not
> an arbitrary file system identified by URI. So we can't put an arbitrary
> fully-qualified URI into yarn.nodemanager.remote-app-log-dir.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
