Github user tgravescs commented on a diff in the pull request:
https://github.com/apache/spark/pull/921#discussion_r13396177
--- Diff:
yarn/stable/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala
---
@@ -256,14 +256,22 @@ class ApplicationMaster(args:
ApplicationMasterArguments, conf: Configuration,
// TODO: Handle container failure
yarnAllocator.addResourceRequests(args.numExecutors)
// Exits the loop if the user thread exits.
- while (yarnAllocator.getNumExecutorsRunning < args.numExecutors &&
userThread.isAlive) {
+ val startTime = System.currentTimeMillis()
+ var usedTime = 0L
+ while ((yarnAllocator.getNumExecutorsRunning < args.numExecutors
+ && userThread.isAlive) && (usedTime < 1000L * 60 * 10)) {
if (yarnAllocator.getNumExecutorsFailed >= maxNumExecutorFailures)
{
finishApplicationMaster(FinalApplicationStatus.FAILED,
"max number of executor failures reached")
}
yarnAllocator.allocateResources()
+ val numExecutorsFailed = yarnAllocator.getNumExecutorsFailed
+ if (numExecutorsFailed > 0) {
+ yarnAllocator.addResourceRequests(numExecutorsFailed)
--- End diff --
I don't think you want to add just the numExecutorsFailed as that could be
way more then what you need. You want to take the difference between what you
want and what is missing and not pending. Similar to how launchReporterThread
does it.
---
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.
---
