Github user kayousterhout commented on a diff in the pull request:
https://github.com/apache/spark/pull/892#discussion_r13414402
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala ---
@@ -195,7 +196,7 @@ private[spark] class TaskSetManager(
}
}
- if (!hadAliveLocations) {
+ if (tasks(index).preferredLocations.isEmpty || (!delaySchedule &&
!hadAliveLocations)) {
--- End diff --
Consistent with what @mateiz said, I think it would be better to have
"delaySchedule" on all the time (so if new executors come up, we always add
them to the tasks' locality lists). I think this line should be the original
version though (just "if (!hadAliveLocations)") -- otherwise if a task really
doesn't have any alive locations (and no more are going to be added in the
future), the task will be stuck waiting to be scheduled until the locality wait
threshold expires, which could be a long time.
---
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.
---
