Github user andrewor14 commented on a diff in the pull request: https://github.com/apache/spark/pull/3607#discussion_r21553460 --- Diff: yarn/common/src/main/scala/org/apache/spark/deploy/yarn/ClientArguments.scala --- @@ -39,23 +39,34 @@ private[spark] class ClientArguments(args: Array[String], sparkConf: SparkConf) var appName: String = "Spark" var priority = 0 - // Additional memory to allocate to containers - // For now, use driver's memory overhead as our AM container's memory overhead - val amMemoryOverhead = sparkConf.getInt("spark.yarn.driver.memoryOverhead", - math.max((MEMORY_OVERHEAD_FACTOR * amMemory).toInt, MEMORY_OVERHEAD_MIN)) - - val executorMemoryOverhead = sparkConf.getInt("spark.yarn.executor.memoryOverhead", - math.max((MEMORY_OVERHEAD_FACTOR * executorMemory).toInt, MEMORY_OVERHEAD_MIN)) - private val isDynamicAllocationEnabled = sparkConf.getBoolean("spark.dynamicAllocation.enabled", false) parseArgs(args.toList) + + val isClusterMode = userClass != null + loadEnvironmentArgs() validateArgs() + // Additional memory to allocate to containers. In different modes, we use different configs. + val amMemOverheadConf = if (isClusterMode) { + "spark.yarn.driver.memoryOverhead" + } else { + "spark.yarn.am.memoryOverhead" --- End diff -- Hm... I agree that adding `clientmode` is clearer but it's also incredibly verbose. I think if we document this clearly that this is only for client mode then the user won't arbitrarily try to set it. I'll comment on #3409 about this as well.
--- 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