[
https://issues.apache.org/jira/browse/SPARK-26395?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16766547#comment-16766547
]
Marcelo Vanzin commented on SPARK-26395:
----------------------------------------
The code that cleans up stages does clean up the RDD graphs:
{noformat}
if (!hasMoreAttempts) {
kvstore.delete(classOf[RDDOperationGraphWrapper], s.info.stageId)
}
{noformat}
Are you sure stages are being properly cleaned up in your case? SPARK-25837
could cause stage cleanup to be really slow, that will be fixed in 2.3.3.
> Spark Thrift server memory leak
> -------------------------------
>
> Key: SPARK-26395
> URL: https://issues.apache.org/jira/browse/SPARK-26395
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 2.3.2
> Reporter: Konstantinos Andrikopoulos
> Priority: Major
>
> We are running Thrift Server in standalone mode and we have observed that the
> heap of the driver is constantly increasing. After analysing the heap dump
> the issue seems to be that the ElementTrackingStore is constantly increasing
> due to the addition of RDDOperationGraphWrapper objects that are not cleaned
> up.
> The ElementTrackingStore defines the addTrigger method were you are able to
> set thresholds in order to perform cleanup but in practice it is used for
> ExecutorSummaryWrapper, JobDataWrapper and StageDataWrapper classes by using
> the following spark properties
> * spark.ui.retainedDeadExecutors
> * spark.ui.retainedJobs
> * spark.ui.retainedStages
> So the RDDOperationGraphWrapper which is been added using the onJobStart
> method of AppStatusListener class [kvstore.write(uigraph) #line 291]
> in not cleaned up and it constantly increases causing a memory leak
--
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
