[
https://issues.apache.org/jira/browse/HIVE-12538?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15038609#comment-15038609
]
Xuefu Zhang commented on HIVE-12538:
------------------------------------
My understanding is that a hive session may "own" more than one spark session
in case of asynchronous queries. If a spark session is live (used to run a
spark job), that spark session will not be used to run the next job. Therefore,
whenever whenever a spark configuration change is detected in Hive session, we
need to mark all the live Spark sessions as outdated. When we are getting a
session from the pool and check if the flag is set, then we destroy it and get
a new one. Hope this will simply things a bit.
> After set spark related config, SparkSession never get reused
> -------------------------------------------------------------
>
> Key: HIVE-12538
> URL: https://issues.apache.org/jira/browse/HIVE-12538
> Project: Hive
> Issue Type: Bug
> Components: Spark
> Affects Versions: 1.3.0
> Reporter: Nemon Lou
> Assignee: Nemon Lou
> Attachments: HIVE-12538.1.patch, HIVE-12538.2.patch,
> HIVE-12538.3.patch, HIVE-12538.4.patch, HIVE-12538.patch
>
>
> Hive on Spark yarn-cluster mode.
> After setting "set spark.yarn.queue=QueueA;" ,
> run the query "select count(*) from test" 3 times and you will find 3
> different yarn applications.
> Two of the yarn applications in FINISHED & SUCCEEDED state,and one in RUNNING
> & UNDEFINED state waiting for next work.
> And if you submit one more "select count(*) from test" ,the third one will be
> in FINISHED & SUCCEEDED state and a new yarn application will start up.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
