[
https://issues.apache.org/jira/browse/YARN-4772?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15185138#comment-15185138
]
Steve Loughran commented on YARN-4772:
--------------------------------------
Assuming this is teardown, the root cause is that the executors only check for
service stop after completing each scan of the FS, which is presumably
{{O(active-apps) * O(fresh-data)}}, with the fresh data time being driven by
cpu and memory requirements of parsing JSON.
Teardown can be addressed by
# having the iterator across files probe for the service being stopped on every
iteration.
# maybe: having the JSON parser check the status in {{LogInfo.doParse()}}
between parsing each element. That could be hidden by having the inner input
stream do the check and raise an EOF on the closure
> Overloaded leveljb can crash the ATS "pthread lock: Invalid argument"
> ---------------------------------------------------------------------
>
> Key: YARN-4772
> URL: https://issues.apache.org/jira/browse/YARN-4772
> Project: Hadoop YARN
> Issue Type: Bug
> Components: timelineserver
> Affects Versions: 2.8.0
> Environment: OSX, scala history scale tests; Java 1.7.0_75-b13
> Reporter: Steve Loughran
> Priority: Minor
>
> while running scale tests with a few hundred thousand events attached to a
> single timeline entity, the JVM crashed
> {code}
> pthread lock: Invalid argument
> /bin/sh: line 1: 8196 Abort trap: 6
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
