[
https://issues.apache.org/jira/browse/HIVE-16433?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15968046#comment-15968046
]
zhihai xu commented on HIVE-16433:
----------------------------------
Thanks for the review [~jxiang]! Currently I didn't have the stack trace, but
it is a potential race condition since user can cancel the query at any time,
ExecDriver.shutdown may be called at any time to set variable {{rj}} to null.
Yes, I plan to use isTaskShutdown function from hook to find out whether
ExecDriver is shutdown by user, so we can better monitor the query from outside.
> Not nullify variable "rj" to avoid NPE due to race condition in ExecDriver.
> ---------------------------------------------------------------------------
>
> Key: HIVE-16433
> URL: https://issues.apache.org/jira/browse/HIVE-16433
> Project: Hive
> Issue Type: Bug
> Reporter: zhihai xu
> Assignee: zhihai xu
> Priority: Minor
> Attachments: HIVE-16433.000.patch
>
>
> Not nullify variable {{rj}} to avoid NPE due to race condition in ExecDriver.
> currently {{rj}} is set to null in ExecDriver.shutdown which is called from
> other thread for query cancellation. It can happen at any time. There is a
> potential race condition, the {{rj}} is still accessed after shutdown is
> called. For example: if the following code is executed right after
> ExecDriver.shutdown is called.
> {code}
> this.jobID = rj.getJobID();
> updateStatusInQueryDisplay();
> returnVal = jobExecHelper.progress(rj, jc, ctx);
> {code}
> Currently the purpose of nullifying {{rj}} is mainly to make sure
> {{rj.killJob()}} is only called once.
> I will add a flag {{jobKilled}} to make sure {{rj.killJob()}} will be only
> called once.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
