[ 
https://issues.apache.org/jira/browse/DRILL-6325?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16434693#comment-16434693
 ] 

Timothy Farkas commented on DRILL-6325:
---------------------------------------

There are also data structures accessed by multiple threads that are not 
synchronized in QueryManager

> Memory usage stats may not be updated correctly for an operator.
> ----------------------------------------------------------------
>
>                 Key: DRILL-6325
>                 URL: https://issues.apache.org/jira/browse/DRILL-6325
>             Project: Apache Drill
>          Issue Type: Bug
>            Reporter: Timothy Farkas
>            Assignee: Timothy Farkas
>            Priority: Major
>
> The memory usage stats for an operator in a query profile may not be up to 
> date, even after the query is completed. This is caused by two possible 
> issues:
> 1 Memory usage is polled by a thread once every 5 seconds. However the final 
> memory usage of the operator may not be reported if the operator completes 
> before the next polling cycle.
> 2 There is a race condition in the Foreman where one thread sets the metrics 
> on an object and another thread reads the metrics object to include in a 
> query profile. The thread that reads the metrics object may read old metrics 
> since access to the variable is not synchronized.
> There are two fixes for this
> 1 Poll the stats one more time after the fragment root completes but before 
> the operator is closed.
> 2 Synchronize the methods on the FragmentData class.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to