[jira] [Commented] (HADOOP-9504) create metricsRateAttributeMod with ConcurrentHashMap to avoid hashmap concurrent race

["Jason Lowe (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HADOOP-9504?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13641880#comment-13641880 ] 

Jason Lowe commented on HADOOP-9504:
------------------------------------

Is this really the right fix?  As I understand it, the problem was caused by two or more threads racing in createMBeanInfo() and that's what clobbered a HashMap causing an infinite loop on a subsequent get.  I agree that making it a ConcurrentHashMap will prevent the infinite loop, but will the resulting map contain what we want?  As is, it could contain a hodgepodge of puts from different threads which seems bad.

createMBeanInfo() creates a fresh map upon each invocation, and I think the real issue is that it makes the new map available to other threads via metricsRateAttributeMod *before* the map has finished being initialized.  I think we should build the map into a local variable then advertise it only after it has been completely filled in.  That way metricsRateAttributeMod is always referring to a consistent map that isn't being actively modified.  The only writes to the map are in createMBeanInfo(), and therefore we may not need a ConcurrentHashMap in that case.
                
> create metricsRateAttributeMod with ConcurrentHashMap to avoid hashmap concurrent race
> --------------------------------------------------------------------------------------
>
>                 Key: HADOOP-9504
>                 URL: https://issues.apache.org/jira/browse/HADOOP-9504
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: metrics
>    Affects Versions: 3.0.0, 2.0.3-alpha
>            Reporter: Liang Xie
>            Assignee: Liang Xie
>            Priority: Critical
>         Attachments: HADOOP-9504.txt
>
>
> Please see HBASE-8416 for detail information.
> we need to take care of the synchronization for HashMap put(), otherwise it may lead to spin loop.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira