[
https://issues.apache.org/jira/browse/HIVE-4574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13815087#comment-13815087
]
Ian Robertson commented on HIVE-4574:
-------------------------------------
I'm seeing this issue in CDH 5 beta 1, which runs jdk1.7.0_25-cloudera and
hive-0.11. I have verified it is the same issue - we're seeing stack traces
similar to the above in three threads across multiple thread dumps, and the
heap dump shows a corrupted WeakHashMap (owned by a WeakCache) with a circular
reference in the linked list for one of its buckets.
Notably, CDH5b1 also has a backported patch of HIVE-1511 - see
http://archive.cloudera.com/cdh5/cdh/5/hive-0.11.0-cdh5.0.0-beta-1.CHANGES.txt.
Perhaps this needs an additional fix?
> XMLEncoder thread safety issues in openjdk7 causes HiveServer2 to be stuck
> --------------------------------------------------------------------------
>
> Key: HIVE-4574
> URL: https://issues.apache.org/jira/browse/HIVE-4574
> Project: Hive
> Issue Type: Bug
> Components: HiveServer2
> Affects Versions: 0.11.0
> Reporter: Thejas M Nair
> Assignee: Thejas M Nair
> Attachments: HIVE-4574.1.patch
>
>
> In open jdk7, XMLEncoder.writeObject call leads to calls to
> java.beans.MethodFinder.findMethod(). MethodFinder class not thread safe
> because it uses a static WeakHashMap that would get used from multiple
> threads. See -
> http://grepcode.com/file/repository.grepcode.com/java/root/jdk/openjdk/7-b147/com/sun/beans/finder/MethodFinder.java#46
> Concurrent access to HashMap implementation that are not thread safe can
> sometimes result in infinite-loops and other problems. If jdk7 is in use, it
> makes sense to synchronize calls to XMLEncoder.writeObject .
--
This message was sent by Atlassian JIRA
(v6.1#6144)
