[
https://issues.apache.org/jira/browse/HIVE-9258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14274528#comment-14274528
]
Xuefu Zhang commented on HIVE-9258:
-----------------------------------
[~jxiang], thanks for looking into this. Looking at the code, I see that it
uses SparkSession instance which is indeed shared with regular queries. Since
this is confirmed, please close this as "not a problem".
BTW, I noticed that we have local cache for sparkMemoryAndCores as in:
{code}
if (sparkMemoryAndCores == null) {
{code}
This would mean that we wouldn't update the value in the entire user session.
However, this value can change dynamically. Do you think we should not cache
the value?
> Explain query shouldn't launch a Spark application [Spark Branch]
> -----------------------------------------------------------------
>
> Key: HIVE-9258
> URL: https://issues.apache.org/jira/browse/HIVE-9258
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Xuefu Zhang
> Assignee: Jimmy Xiang
>
> Currently for Hive on Spark, query plan includes the number of reducers,
> which is determined partly by the Spark cluster. Thus, explain query will
> need to launch a Spark application (Spark remote context), which is costly.
> To make things worse, the application is discarded right way.
> Ideally, we shouldn't launch a Spark application even for an explain query.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
