Github user andrewor14 commented on a diff in the pull request: https://github.com/apache/spark/pull/6430#discussion_r31278378 --- Diff: core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala --- @@ -262,15 +267,22 @@ private[spark] class ExecutorAllocationManager( val maxNeeded = maxNumExecutorsNeeded if (maxNeeded < numExecutorsTarget) { - // The target number exceeds the number we actually need, so stop adding new - // executors and inform the cluster manager to cancel the extra pending requests - val oldNumExecutorsTarget = numExecutorsTarget - numExecutorsTarget = math.max(maxNeeded, minNumExecutors) - client.requestTotalExecutors(numExecutorsTarget) - numExecutorsToAdd = 1 - logInfo(s"Lowering target number of executors to $numExecutorsTarget because " + - s"not all requests are actually needed (previously $oldNumExecutorsTarget)") - numExecutorsTarget - oldNumExecutorsTarget + if (!numTargetExecutorAdjustable.get) { + // Keep the initial number of target executor to not ramp down until the first job is + // submitted or the first idle executor is released. + client.requestTotalExecutors(numExecutorsTarget) --- End diff -- Actually, does it make sense to do this inside the case `maxNeeded < numExecutorsTarget`? This condition implies that we are targeting more executors that are needed, and so we should cancel outstanding requests. This is not really true during initialization. I think a better indication that your application has initialized is if the number of executors is >= your initial executors.
--- 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. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org