[
https://issues.apache.org/jira/browse/SPARK-23670?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16606151#comment-16606151
]
Marcelo Vanzin commented on SPARK-23670:
----------------------------------------
How many instances of the following classes do you have in your dump:
- {{SQLExecutionUIData}}
- {{SparkPlanGraphWrapper}}
- {{SparkPlanGraphNodeWrapper}}
With the fix I'd expect the first two to be capped at 1000 (unless you changed
the config), but the latter may still be high if you have complicated queries.
That could explain high memory usage.
> Memory leak of SparkPlanGraphWrapper in sparkUI
> -----------------------------------------------
>
> Key: SPARK-23670
> URL: https://issues.apache.org/jira/browse/SPARK-23670
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.3.0
> Reporter: Myroslav Lisniak
> Assignee: Myroslav Lisniak
> Priority: Major
> Fix For: 2.3.1, 2.4.0
>
> Attachments: heap.png, heapdump_OOM.png
>
>
> Memory leak on driver for a long time running application. We have
> application using structured streaming and running 48 hours. But driver fails
> with out of memory after 25 hours. After investigating heap dump we found
> that most of the memory was occupied with a lot of *SparkPlanGraphWrapper*
> objects inside *InMemoryStore*.
> Application was run with option:
> --driver-memory 4G
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
