Github user tdas commented on a diff in the pull request:
https://github.com/apache/spark/pull/8340#discussion_r37713192
--- Diff:
streaming/src/main/scala/org/apache/spark/streaming/scheduler/ReceiverTracker.scala
---
@@ -431,7 +450,8 @@ class ReceiverTracker(ssc: StreamingContext,
skipReceiverLaunch: Boolean = false
receiver.preferredLocation,
receiverTrackingInfos,
getExecutors)
- updateReceiverScheduledExecutors(receiver.streamId,
scheduledExecutors)
--- End diff --
@zsxwing explained me this offline. In case of resecheduling, the scheduled
executors are not stored so that the following scenario does not occur.
- Initial globally-optimal schedule is stored, but one receiver gets
launched incorrectly.
- The receiver is rejected and therefore has to be rescheduled, but if the
rescheduled location (which is locally-optimal for that receiver) is saved, it
will overwrite the original globally optimal location, and will get launched
somewhere else that does not ensure the proper global balancing.
---
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]
