[
https://issues.apache.org/jira/browse/YARN-4773?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15187250#comment-15187250
]
Jun Gong commented on YARN-4773:
--------------------------------
Thanks [~jlowe] for reporting the issue.
YARN-4720 tries to skip unnecessary NN operations for every call
*AppLogAggregatorImpl#uploadLogsForContainers* if pendingContainerInThisCycle
is empty. IIUC [~jlowe] means this case: rolling log aggregation is disabled,
when app completes, we call
*AppLogAggregatorImpl#uploadLogsForContainers(true)*, and
*pendingContainerInThisCycle* is not empty, then we will call
*AppLogAggregatorImpl#cleanOldLogs*, however we do not need call
*AppLogAggregatorImpl#cleanOldLogs* because there have been no containers' logs
uploaded before.
> Log aggregation performs extraneous filesystem operations when rolling log
> aggregation is disabled
> --------------------------------------------------------------------------------------------------
>
> Key: YARN-4773
> URL: https://issues.apache.org/jira/browse/YARN-4773
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 2.6.0
> Reporter: Jason Lowe
> Priority: Minor
>
> I noticed when log aggregation occurs for an application the nodemanager is
> listing the application's log directory in HDFS. Apparently this is for
> removing old logs before uploading new ones. This is a wasteful operation
> when rolling log aggregation is disabled, since there will be no prior logs
> in HDFS -- aggregation only occurs once when rolling log aggregation is
> disabled.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
