Github user kayousterhout commented on a diff in the pull request:
https://github.com/apache/spark/pull/15986#discussion_r89414663
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/TaskSchedulerImpl.scala ---
@@ -525,7 +527,14 @@ private[spark] class TaskSchedulerImpl(
* of any running tasks, since the loss reason defines whether we'll
fail those tasks.
*/
private def removeExecutor(executorId: String, reason:
ExecutorLossReason) {
- executorIdToTaskCount -= executorId
+ executorIdToRunningTaskIds.remove(executorId).foreach { taskIds =>
+ logDebug("Cleaning up TaskScheduler state for tasks " +
+ s"${taskIds.mkString("[", ",", "]")} on failed executor
$executorId")
+ taskIds.foreach { tid =>
+ taskIdToExecutorId.remove(tid)
+ taskIdToTaskSetManager.remove(tid)
--- End diff --
What about moving the two above lines to a "cleanupTaskState" method (or
something like that) that can be called here and in statusUpdate? I wonder if
that will help us avoid state tracking issues in the future -- because it's
clear that the same cleanup should be happening in both places.
Can you also add a comment here saying that we only need to cleanup state
because telling the TSM that the task has failed happens via the
rootPool.executorLost call?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
