[jira] [Updated] (SPARK-12920) Spark thrift server can run at very high CPU with concurrent users

["Rajesh Balamohan (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/SPARK-12920?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rajesh Balamohan updated SPARK-12920:
-------------------------------------
    Attachment: SPARK-12920.profiler.png
                SPARK-12920.profiler_job_progress_listner.png

-RDD would be different if user runs different queries.
-This causes many objects in cache causing lots of pressure for GC (Refer 2703 MB of cache size).
-Also, JobProgressListener caches stageInfo and Job objects.  Setting spark.sql.ui.retainedExecutions=0, spark.ui.enabled=false, spark.ui.retainedStages=0, spark.ui.retainedJobs=0 did not release the objects . This is because of https://github.com/apache/spark/blob/2b5d11f34d73eb7117c0c4668c1abb27dcc3a403/core/src/main/scala/org/apache/spark/ui/jobs/JobProgressListener.scala#L142 (releases one entry when retainedStages=0.  This causes memory build up in multi user env).




> Spark thrift server can run at very high CPU with concurrent users
> ------------------------------------------------------------------
>
>                 Key: SPARK-12920
>                 URL: https://issues.apache.org/jira/browse/SPARK-12920
>             Project: Spark
>          Issue Type: Improvement
>          Components: SQL
>            Reporter: Rajesh Balamohan
>         Attachments: SPARK-12920.profiler.png, SPARK-12920.profiler_job_progress_listner.png
>
>
> - Configured with fair-share-scheduler.
> - 4-5 users submitting/running jobs concurrently via spark-thrift-server
> - Spark thrift server spikes to1600+% CPU and stays there for long time



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org