[
https://issues.apache.org/jira/browse/HIVE-21035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16719410#comment-16719410
]
Antal Sinkovits commented on HIVE-21035:
----------------------------------------
[~xuefuz] Thanks for checking on this.
The application master starts up, when the first query is executed. If two
queries are executed simultaneously, (and the AM is not running) , or the AM
starts up slow, and the other query is executed during its startup, then it can
happen that both queries try to kick one off. Unfortunately when this happens,
one of them wont get shutdown.
I was able two reproduce the issue two way. Once with HUE, which can execute
multiple queries simultaneously in one session.
Also programatically using the same JDBC connection with multiple threads
creating their own statements.
> Race condition in SparkUtilities#getSparkSession
> ------------------------------------------------
>
> Key: HIVE-21035
> URL: https://issues.apache.org/jira/browse/HIVE-21035
> Project: Hive
> Issue Type: Bug
> Components: Spark
> Affects Versions: 4.0.0
> Reporter: Antal Sinkovits
> Assignee: Antal Sinkovits
> Priority: Major
> Attachments: HIVE-21035.01.patch
>
>
> It can happen, that when in one given session, multiple queries are executed,
> that due to a race condition, multiple spark application master gets kicked
> off.
> In this case, the one that started earlier, will not be killed, when the hive
> session closes, consuming resources.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
