Github user vanzin commented on a diff in the pull request:
https://github.com/apache/spark/pull/14926#discussion_r77720376
--- Diff:
core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala ---
@@ -392,10 +397,36 @@ private[spark] class ExecutorAllocationManager(
}
/**
- * Request the cluster manager to remove the given executor.
+ * Request the cluster manager to remove the given executors.
* Return whether the request is received.
*/
- private def removeExecutor(executorId: String): Boolean = synchronized {
+ private def removeExecutors(executorIds: Seq[String]): Boolean =
synchronized {
+
+ val executorIdsToBeRemoved = executorIds.filter(canBeKilled)
+
+ // Send a request to the backend to kill this executor
+ val removeRequestAcknowledged = testing ||
client.killExecutors(executorIdsToBeRemoved)
--- End diff --
The problem is not the backend. The problem is
`CoarseGrainedSchedulerBackend.killExecutors` which can decide to kill just a
subset of the executors asked to be killed.
I'm wary of @dhruve's assessment of the problem because that code was added
*exactly* because the code was killing busy executors that had for some reason
become "idle". See SPARK-9552.
---
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]
