Github user dragos commented on a diff in the pull request:
https://github.com/apache/spark/pull/10924#discussion_r53029744
--- Diff:
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/CoarseMesosSchedulerBackend.scala
---
@@ -254,53 +258,65 @@ private[spark] class CoarseMesosSchedulerBackend(
val cpus = getResource(offer.getResourcesList, "cpus").toInt
val id = offer.getId.getValue
if (meetsConstraints) {
- if (taskIdToSlaveId.size < executorLimit &&
- totalCoresAcquired < maxCores &&
- mem >= calculateTotalMemory(sc) &&
- cpus >= 1 &&
- failuresBySlaveId.getOrElse(slaveId, 0) < MAX_SLAVE_FAILURES
&&
- !slaveIdsWithExecutors.contains(slaveId)) {
- // Launch an executor on the slave
- val cpusToUse = math.min(cpus, maxCores - totalCoresAcquired)
- totalCoresAcquired += cpusToUse
- val taskId = newMesosTaskId()
- taskIdToSlaveId.put(taskId, slaveId)
- slaveIdsWithExecutors += slaveId
- coresByTaskId(taskId) = cpusToUse
- // Gather cpu resources from the available resources and use
them in the task.
- val (remainingResources, cpuResourcesToUse) =
- partitionResources(offer.getResourcesList, "cpus", cpusToUse)
- val (_, memResourcesToUse) =
- partitionResources(remainingResources.asJava, "mem",
calculateTotalMemory(sc))
- val taskBuilder = MesosTaskInfo.newBuilder()
-
.setTaskId(TaskID.newBuilder().setValue(taskId.toString).build())
- .setSlaveId(offer.getSlaveId)
- .setCommand(createCommand(offer, cpusToUse +
extraCoresPerSlave, 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())
+ if (totalCoresAcquired < maxCores) {
+ if (taskIdToSlaveId.size < executorLimit &&
+ mem >= calculateTotalMemory(sc) &&
+ cpus >= 1 &&
+ failuresBySlaveId.getOrElse(slaveId, 0) <
MAX_SLAVE_FAILURES &&
+ !slaveIdsWithExecutors.contains(slaveId)) {
+ // Launch an executor on the slave
+ val cpusToUse = math.min(cpus, maxCores - totalCoresAcquired)
+ totalCoresAcquired += cpusToUse
+ val taskId = newMesosTaskId()
+ taskIdToSlaveId.put(taskId, slaveId)
+ slaveIdsWithExecutors += slaveId
+ coresByTaskId(taskId) = cpusToUse
+ // Gather cpu resources from the available resources and use
them in the task.
+ val (remainingResources, cpuResourcesToUse) =
+ partitionResources(offer.getResourcesList, "cpus",
cpusToUse)
+ val (_, memResourcesToUse) =
+ partitionResources(remainingResources.asJava, "mem",
calculateTotalMemory(sc))
+ val taskBuilder = MesosTaskInfo.newBuilder()
+
.setTaskId(TaskID.newBuilder().setValue(taskId.toString).build())
+ .setSlaveId(offer.getSlaveId)
+ .setCommand(createCommand(offer, cpusToUse +
extraCoresPerSlave, 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())
+ }
+
+ // Accept the offer and launch the task
+ logDebug(s"Accepting offer: $id with attributes:
$offerAttributes" +
+ s" mem: $mem cpu: $cpus")
+ slaveIdToHost(offer.getSlaveId.getValue) = offer.getHostname
+ d.launchTasks(
+ Collections.singleton(offer.getId),
+ Collections.singleton(taskBuilder.build()), filters)
+ } else {
+ // Decline the offer
+ logDebug(s"Declining offer: $id with attributes:
$offerAttributes" +
+ s" mem: $mem cpu: $cpus")
+ d.declineOffer(offer.getId)
}
-
- // Accept the offer and launch the task
- logDebug(s"Accepting offer: $id with attributes:
$offerAttributes mem: $mem cpu: $cpus")
- slaveIdToHost(offer.getSlaveId.getValue) = offer.getHostname
- d.launchTasks(
- Collections.singleton(offer.getId),
- Collections.singleton(taskBuilder.build()), filters)
} else {
- // Decline the offer
- logDebug(s"Declining offer: $id with attributes:
$offerAttributes mem: $mem cpu: $cpus")
- d.declineOffer(offer.getId)
+ // We reached the maximum number of cores for this framework.
We don't need to see
+ // new offers. Decline the offer for a long period of time.
+ logDebug(s"Declining offer (reached max cores): $id with
attributes:" +
+ s" $offerAttributes mem: $mem cpu: $cpus" +
+ s" for $rejectOfferDurationForReachedMaxCores seconds")
+ d.declineOffer(offer.getId, Filters.newBuilder()
+
.setRefuseSeconds(rejectOfferDurationForReachedMaxCores).build())
--- End diff --
I think the two cases where offers are rejected for a longer period should
be consolidated in a simple helper function that logs the reason why and
declines the offer. I'd also reorganize the code to be less nested:
```
if (!meetsConstraints) {
declineFor("unmet constraints", rejectOfferDurationUnmetConstraints)
} else if (totalCoresAcquired >= maxCores) {
declineFor("reached max cores", rejectOfferDurationMaxCores)
} else {
.. happy case
}
```
---
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]
