Github user JoshRosen commented on a diff in the pull request:
https://github.com/apache/spark/pull/3638#discussion_r22202222
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/TaskSchedulerImpl.scala ---
@@ -256,15 +256,21 @@ private[spark] class TaskSchedulerImpl(
val execId = shuffledOffers(i).executorId
val host = shuffledOffers(i).host
if (availableCpus(i) >= CPUS_PER_TASK) {
- for (task <- taskSet.resourceOffer(execId, host, maxLocality))
{
- tasks(i) += task
- val tid = task.taskId
- taskIdToTaskSetId(tid) = taskSet.taskSet.id
- taskIdToExecutorId(tid) = execId
- executorsByHost(host) += execId
- availableCpus(i) -= CPUS_PER_TASK
- assert(availableCpus(i) >= 0)
- launchedTask = true
+ try {
+ for (task <- taskSet.resourceOffer(execId, host,
maxLocality)) {
+ tasks(i) += task
+ val tid = task.taskId
+ taskIdToTaskSetId(tid) = taskSet.taskSet.id
+ taskIdToExecutorId(tid) = execId
+ executorsByHost(host) += execId
+ availableCpus(i) -= CPUS_PER_TASK
+ assert(availableCpus(i) >= 0)
+ launchedTask = true
+ }
+ } catch {
+ case e: TaskNotSerializableException => {
--- End diff --
Actually, I don't think that returning an empty sequence is the right call
here, since even though this _particular_ task set might have failed to launch
a task, we might still have task sets that can be launched. So, it seems like
we'd like to break out of the innermost loop rather than returning from
`resourceOffers`. To do this, it might make sense to split these nested loops
into a pair of functions where we can `return` from the "launch tasks from this
particular task set" function.
---
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]
