Github user drcrallen commented on a diff in the pull request:
https://github.com/apache/spark/pull/13715#discussion_r74952999
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosCoarseGrainedSchedulerBackend.scala
---
@@ -382,59 +382,97 @@ private[spark] class
MesosCoarseGrainedSchedulerBackend(
for (offer <- offers) {
val slaveId = offer.getSlaveId.getValue
val offerId = offer.getId.getValue
- val resources = remainingResources(offerId)
-
- if (canLaunchTask(slaveId, resources)) {
- // Create a task
- launchTasks = true
- val taskId = newMesosTaskId()
- val offerCPUs = getResource(resources, "cpus").toInt
-
- val taskCPUs = executorCores(offerCPUs)
- val taskMemory = executorMemory(sc)
-
- slaves.getOrElseUpdate(slaveId, new
Slave(offer.getHostname)).taskIDs.add(taskId)
-
- val (afterCPUResources, cpuResourcesToUse) =
- partitionResources(resources, "cpus", taskCPUs)
- val (resourcesLeft, memResourcesToUse) =
- partitionResources(afterCPUResources.asJava, "mem", taskMemory)
-
- val taskBuilder = MesosTaskInfo.newBuilder()
-
.setTaskId(TaskID.newBuilder().setValue(taskId.toString).build())
- .setSlaveId(offer.getSlaveId)
- .setCommand(createCommand(offer, taskCPUs +
extraCoresPerExecutor, taskId))
- .setName("Task " + taskId)
- .addAllResources(cpuResourcesToUse.asJava)
- .addAllResources(memResourcesToUse.asJava)
-
- sc.conf.getOption("spark.mesos.executor.docker.image").foreach {
image =>
- MesosSchedulerBackendUtil
- .setupContainerBuilderDockerInfo(image, sc.conf,
taskBuilder.getContainerBuilder)
+ val availableResources = remainingResources(offerId)
+ val offerMem = getResource(availableResources, "mem")
+ val offerCpu = getResource(availableResources, "cpus")
+
+ // Catch offer limits
+ calculateUsableResources(
+ sc,
+ offerCpu.toInt,
+ offerMem.toInt
+ ).flatMap(
+ {
+ // Catch "global" limits
+ case (taskCPUs: Int, taskMemory: Int) =>
+ if (numExecutors() >= executorLimit) {
+ logTrace(s"${numExecutors()} exceeds limit of
$executorLimit")
+ None
+ } else if (
+ slaves.get(slaveId).map(_.taskFailures).getOrElse(0) >=
MAX_SLAVE_FAILURES
+ ) {
+ logTrace(s"Slave $slaveId exceeded limit of
$MAX_SLAVE_FAILURES failures")
+ None
+ } else {
+ Some((taskCPUs, taskMemory))
+ }
}
-
- tasks(offer.getId) ::= taskBuilder.build()
- remainingResources(offerId) = resourcesLeft.asJava
- totalCoresAcquired += taskCPUs
- coresByTaskId(taskId) = taskCPUs
+ ) match {
+ case Some((taskCPUs: Int, taskMemory: Int)) =>
+ // Create a task
+ launchTasks = true
+ val taskId = newMesosTaskId()
+
+ slaves.getOrElseUpdate(slaveId, new
Slave(offer.getHostname)).taskIDs.add(taskId)
+
+ val (afterCPUResources, cpuResourcesToUse) =
+ partitionResources(availableResources, "cpus", taskCPUs)
+ val (resourcesLeft, memResourcesToUse) =
+ partitionResources(afterCPUResources.asJava, "mem",
taskMemory)
+
+ val taskBuilder = MesosTaskInfo.newBuilder()
+
.setTaskId(TaskID.newBuilder().setValue(taskId.toString).build())
+ .setSlaveId(offer.getSlaveId)
+ .setCommand(createCommand(offer, taskCPUs +
extraCoresPerExecutor, taskId))
+ .setName("Task " + taskId)
+ .addAllResources(cpuResourcesToUse.asJava)
+ .addAllResources(memResourcesToUse.asJava)
+
+ sc.conf.getOption("spark.mesos.executor.docker.image").foreach
{ image =>
+ MesosSchedulerBackendUtil
+ .setupContainerBuilderDockerInfo(image, sc.conf,
taskBuilder.getContainerBuilder)
+ }
+
+ tasks(offer.getId) ::= taskBuilder.build()
+ remainingResources(offerId) = resourcesLeft.asJava
+ totalCoresAcquired += taskCPUs
+ coresByTaskId(taskId) = taskCPUs
+ case None => logDebug(s"Insufficient offer CPU: $offerCpu MEM:
$offerMem")
}
}
}
tasks.toMap
}
- private def canLaunchTask(slaveId: String, resources: JList[Resource]):
Boolean = {
- val offerMem = getResource(resources, "mem")
- val offerCPUs = getResource(resources, "cpus").toInt
- val cpus = executorCores(offerCPUs)
- val mem = executorMemory(sc)
-
- cpus > 0 &&
- cpus <= offerCPUs &&
- cpus + totalCoresAcquired <= maxCores &&
- mem <= offerMem &&
- numExecutors() < executorLimit &&
- slaves.get(slaveId).map(_.taskFailures).getOrElse(0) <
MAX_SLAVE_FAILURES
+ /**
+ * Try and fit the resources to the constraints. Return None if it does
not match
+ *
+ * @param sc Spark context
+ * @param availableCpus The available CPUs
+ * @param availableMem The available Memory
+ * @return Tuple of CPU (integer cores) and Memory (integer MB) desired
+ */
+ private[mesos]
+ def calculateUsableResources(sc: SparkContext, availableCpus: Int,
availableMem: Int):
+ Option[(Int, Int)] = {
+ val desiredMemory = executorMemory(sc)
+ val desiredCpu = executorCores(availableCpus)
--- End diff --
I can do that. Will be a few weeks before I can tackle this again though.
---
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]
