Ngone51 commented on a change in pull request #23677: [SPARK-26755][SCHEDULER]
: Optimize Spark Scheduler to dequeue speculative tasks…
URL: https://github.com/apache/spark/pull/23677#discussion_r305599471
##########
File path: core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
##########
@@ -302,16 +260,23 @@ private[spark] class TaskSetManager(
private def dequeueTaskFromList(
execId: String,
host: String,
- list: ArrayBuffer[Int]): Option[Int] = {
+ list: ArrayBuffer[Int],
+ speculative: Boolean = false): Option[Int] = {
var indexOffset = list.size
while (indexOffset > 0) {
indexOffset -= 1
val index = list(indexOffset)
- if (!isTaskBlacklistedOnExecOrNode(index, execId, host)) {
+ if (!isTaskBlacklistedOnExecOrNode(index, execId, host) &&
+ !(speculative && hasAttemptOnHost(index, host))) {
// This should almost always be list.trimEnd(1) to remove tail
list.remove(indexOffset)
- if (copiesRunning(index) == 0 && !successful(index)) {
- return Some(index)
+ if (!successful(index)) {
+ if (copiesRunning(index) == 0) {
+ return Some(index)
+ } else if (speculative && copiesRunning(index) == 1) {
+ speculatableTasks -= index
Review comment:
when `copiesRunning(index) == 0`, it is still possible to return a
speculatable tasks, thinking about the case that a long running task failed
after it got a pending speculatable task. I mean, for `if
(copiesRunning(index) == 0)` above, it actually covers two cases: 1) not
speculative 2) speculative. So, I think it's better to put this after
`dequeueTaskHelper(execId, host, maxLocality, true)`.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to 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]
