Github user tnachen commented on a diff in the pull request:
https://github.com/apache/spark/pull/4027#discussion_r26988213
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala
---
@@ -204,35 +209,43 @@ private[spark] class CoarseMesosSchedulerBackend(
for (offer <- offers) {
val slaveId = offer.getSlaveId.toString
- val mem = getResource(offer.getResourcesList, "mem")
- val cpus = getResource(offer.getResourcesList, "cpus").toInt
- if (totalCoresAcquired < maxCores &&
- mem >= MemoryUtils.calculateTotalMemory(sc) &&
- cpus >= 1 &&
+ var totalMem = getResource(offer.getResourcesList, "mem")
+ var totalCpus = getResource(offer.getResourcesList, "cpus").toInt
+
+ val tasks = new ArrayBuffer[MesosTaskInfo]
+
+ var executorCount = slaveIdsWithExecutors.getOrElse(slaveId, 0)
+ // Launch as many executors that the resources are available and
less
+ // than the configured max executors per slave.
+ while (totalCoresAcquired < maxCores &&
+ totalMem >= memRequired &&
+ totalCpus >= 1 &&
failuresBySlaveId.getOrElse(slaveId, 0) < MAX_SLAVE_FAILURES &&
- !slaveIdsWithExecutors.contains(slaveId)) {
- // Launch an executor on the slave
- val cpusToUse = math.min(cpus, maxCores - totalCoresAcquired)
+ executorCount < maxExecutorsPerSlave) {
+ val cpusToUse =
+ math.min(maxCpusPerExecutor, math.min(totalCpus, maxCores -
totalCoresAcquired))
totalCoresAcquired += cpusToUse
+ totalCpus -= cpusToUse
+ totalMem -= memRequired
val taskId = newMesosTaskId()
taskIdToSlaveId(taskId) = slaveId
- slaveIdsWithExecutors += slaveId
+ executorCount += 1
+ slaveIdsWithExecutors(slaveId) = executorCount
coresByTaskId(taskId) = cpusToUse
- val task = MesosTaskInfo.newBuilder()
+ tasks.add(MesosTaskInfo.newBuilder()
.setTaskId(TaskID.newBuilder().setValue(taskId.toString).build())
.setSlaveId(offer.getSlaveId)
.setCommand(createCommand(offer, cpusToUse +
extraCoresPerSlave))
.setName("Task " + taskId)
.addResources(createResource("cpus", cpusToUse))
- .addResources(createResource("mem",
- MemoryUtils.calculateTotalMemory(sc)))
- .build()
- d.launchTasks(
- Collections.singleton(offer.getId),
Collections.singletonList(task), filters)
+ .addResources(createResource("mem", memRequired))
+ .build())
+ }
+
+ if (tasks.isEmpty) {
+ d.declineOffer(offer.getId)
--- End diff --
This call is basically equivalent to the old code:
- d.launchTasks(
- Collections.singleton(offer.getId),
Collections.emptyList[MesosTaskInfo](), filters)
It's just that it's much more obvious that we're declining offers, and not
trying to launch empty set of tasks.
---
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
