[
https://issues.apache.org/jira/browse/SPARK-27214?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dongjoon Hyun updated SPARK-27214:
----------------------------------
Affects Version/s: (was: 3.0.0)
3.1.0
> Upgrading locality level when lots of pending tasks have been waiting more
> than locality.wait
> ---------------------------------------------------------------------------------------------
>
> Key: SPARK-27214
> URL: https://issues.apache.org/jira/browse/SPARK-27214
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 3.1.0
> Reporter: liupengcheng
> Priority: Major
>
> Currently, Spark locality wait mechanism is not friendly for large job, when
> number of tasks is large（e.g. 10000+）and with a large number of
> executors(e.g. 2000), executors may be launched on some nodes where the
> locality is not the best(not the same nodes hold HDFS blocks). There are
> cases when `TaskSetManager.lastLaunchTime` is refreshed due to finished tasks
> within `spark.locality.wait` but coming at low rate(e.g. every
> `spark.locality.wait` seconds a task is finished), so locality level would
> not be upgraded and lots of pending tasks will wait a long time.
> In this case, when `spark.dynamicAllocation.enabled=true`, then lots of
> executors may be removed by Driver due to become idle and finally slow down
> the job.
> We encountered this issue in our production spark cluster, it caused lots of
> resources wasting and slowed down user's application.
> Actually, we can optimize this by following formula:
> Suppose numPendingTasks=10000, localityExecutionGainFactor=0.1,
> probabilityOfNextLocalitySchedule=0.5
> {code:java}
> maxTolerableStarvingTime = numTasksCanRun * medianOfTaskExecutionTime *
> localityExecutionGainFactor * probabilityOfNextLocalitySchedule
> totalStarvingTime = sum(starvingTimeByTasks)
> if (totalStarvingTime > maxTolerableStarvingTime)
> { upgrading locality level... }{code}
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
