liupc commented on a change in pull request #23842: [SPARK-26927]Fix race
condition may cause dynamic allocation not working
URL: https://github.com/apache/spark/pull/23842#discussion_r260232554
##########
File path: core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala
##########
@@ -566,6 +574,13 @@ private[spark] class ExecutorAllocationManager(
if (executorIds.contains(executorId)) {
executorIds.remove(executorId)
removeTimes.remove(executorId)
+ if (!removedExecutorIds.offer(executorId)) {
+ logWarning(s"The queue is full for tracking removed executors, will
trim old " +
+ s"executor ids and offer again")
+ val trimSize = math.max(1, removedExecutorIds.size() / 10)
+ (0 to trimSize).foreach(removedExecutorIds.poll())
+ removedExecutorIds.offer(executorId)
Review comment:
@kiszk Yes, I think `ArrayDeque` is ok, and I was thinking if there is a
better solution not based on the following assumption, any other good solution
for this?
> // Assumes the late events for the executor will be correctly handled
before the removal
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
