[
https://issues.apache.org/jira/browse/STORM-1756?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15268592#comment-15268592
]
ASF GitHub Bot commented on STORM-1756:
---------------------------------------
Github user srdo commented on the pull request:
https://github.com/apache/storm/pull/1387#issuecomment-216508276
I did see OOME due to KafkaServer resources. 8x 130MB byte arrays caused
the VM to run out of memory. As I said, I can't reproduce it. It's probably not
an issue on master when storm-kafka gets downgraded to 0.8.2.1 (here
https://github.com/apache/storm/pull/1386), since one of the changes in Kafka
0.9.0.1 was to enable the log cleaner thread by default (and thus allocate some
very large byte arrays when starting a server).
> Storm-kafka tests leak resources due to retained references to KafkaServer
> --------------------------------------------------------------------------
>
> Key: STORM-1756
> URL: https://issues.apache.org/jira/browse/STORM-1756
> Project: Apache Storm
> Issue Type: Bug
> Components: storm-kafka
> Reporter: Stig Rohde Døssing
> Assignee: Stig Rohde Døssing
>
> Since JUnit keeps a reference to each test class until all tests for a module
> have run, the KafkaTestBroker fields are not eligible for garbage collection
> until all tests have run. KafkaTestBroker keeps a reference to
> KafkaServerStartable/KafkaServer, which allocates resources that are also not
> eligible for collection. This very quickly causes surefire to run out of
> memory if this class is used more than a few places.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
