Github user felixcheung commented on a diff in the pull request: https://github.com/apache/spark/pull/21033#discussion_r181627739 --- Diff: resource-managers/mesos/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosCoarseGrainedSchedulerBackend.scala --- @@ -495,9 +500,8 @@ private[spark] class MesosCoarseGrainedSchedulerBackend( launchTasks = true val taskId = newMesosTaskId() val offerCPUs = getResource(resources, "cpus").toInt - val taskGPUs = Math.min( - Math.max(0, maxGpus - totalGpusAcquired), getResource(resources, "gpus").toInt) - + val offerGPUs = getResource(resources, "gpus").toInt + var taskGPUs = executorGpus --- End diff -- so looks like we are changing the behavior for the value set in `spark.mesos.gpus.max` (since 2.1)? we are ok with that/that might break existing deployment? is there a migration guide for something like this? in addition, is there other changes by default - specifically now taskGPUs defaults to 0? also, should we warn if `spark.mesos.executor.gpus` is > `spark.mesos.gpus.max`?
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org