Dongjoon Hyun created SPARK-31645:
-------------------------------------
Summary: Improve ExecutorMonitor shuffle-tracking to handle
speculative tasks properly
Key: SPARK-31645
URL: https://issues.apache.org/jira/browse/SPARK-31645
Project: Spark
Issue Type: Improvement
Components: Spark Core
Affects Versions: 3.1.0
Reporter: Dongjoon Hyun
-
https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/scheduler/dynalloc/ExecutorMonitor.scala#L293-L305
{code:java}
// If the task succeeded and the stage generates shuffle data, record
that this executor
// holds data for the shuffle. This code will track all executors that
generate shuffle
// for the stage, even if speculative tasks generate duplicate shuffle
data and end up
// being ignored by the map output tracker.
//
// This means that an executor may be marked as having shuffle data, and
thus prevented
// from being removed, even though the data may not be used.
if (shuffleTrackingEnabled && event.reason == Success) {
stageToShuffleID.get(event.stageId).foreach { shuffleId =>
exec.addShuffle(shuffleId)
}
}
{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]
