Github user CodingCat commented on a diff in the pull request:
https://github.com/apache/spark/pull/731#discussion_r12511614
--- Diff: core/src/main/scala/org/apache/spark/deploy/master/Master.scala
---
@@ -532,6 +516,99 @@ private[spark] class Master(
}
}
+ private def startMultiExecutorsPerWorker() {
+ // allow user to run multiple executors in the same worker
+ // (within the same worker JVM process)
+ if (spreadOutApps) {
+ for (app <- waitingApps if app.coresLeft > 0) {
+ val memoryPerExecutor = app.desc.memoryPerExecutor
+ var usableWorkers = workers.toArray.filter(_.state ==
WorkerState.ALIVE).
+ filter(worker => worker.coresFree > 0 && worker.memoryFree >=
memoryPerExecutor).
+ sortBy(_.memoryFree / memoryPerExecutor).reverse
+ val maxCoreNumPerExecutor = app.desc.maxCorePerExecutor.get
+ // get the maximum total number of executors we can assign
+ var maxLeftExecutorsToAssign = usableWorkers.map(_.memoryFree /
memoryPerExecutor).sum
+ var maxCoresLeft = maxLeftExecutorsToAssign * maxCoreNumPerExecutor
--- End diff --
the idea here is, user has an expectation on the maximum cores to assign to
the application, but this expectation is usually not achievable due to the
limited cores in each worker;
so the allocation here is to decide executorNum per Worker according to the
memory space on each worker (because this is a hard limitation), and meet the
user's expectation on the cores with the best efforts
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
