Github user tnachen commented on a diff in the pull request:

    https://github.com/apache/spark/pull/9027#discussion_r47868971
  
    --- Diff: 
core/src/main/scala/org/apache/spark/scheduler/cluster/mesos/MesosSchedulerBackend.scala
 ---
    @@ -239,44 +250,69 @@ private[spark] class MesosSchedulerBackend(
               
.setRefuseSeconds(rejectOfferDurationForUnmetConstraints).build())
           }
     
    -      // Of the matching constraints, see which ones give us enough memory 
and cores
    -      val (usableOffers, unUsableOffers) = 
offersMatchingConstraints.partition { o =>
    -        val mem = getResource(o.getResourcesList, "mem")
    -        val cpus = getResource(o.getResourcesList, "cpus")
    -        val slaveId = o.getSlaveId.getValue
    -        val offerAttributes = toAttributeMap(o.getAttributesList)
    -
    -        // check offers for
    -        //  1. Memory requirements
    -        //  2. CPU requirements - need at least 1 for executor, 1 for task
    -        val meetsMemoryRequirements = mem >= calculateTotalMemory(sc)
    -        val meetsCPURequirements = cpus >= (mesosExecutorCores + 
scheduler.CPUS_PER_TASK)
    -        val meetsRequirements =
    -          (meetsMemoryRequirements && meetsCPURequirements) ||
    +    // Of the matching constraints, see which ones give us enough memory 
and cores
    +    val (usableOffers, unUsableOffers) = 
offersMatchingConstraints.partition { o =>
    +      val mem = getResource(o.getResourcesList, "mem")
    +      val cpus = getResource(o.getResourcesList, "cpus")
    +      val slaveId = o.getSlaveId.getValue
    +      val offerAttributes = toAttributeMap(o.getAttributesList)
    +
    +      // check offers for
    +      //  1. Memory requirements
    +      //  2. CPU requirements - need at least 1 for executor, 1 for task
    +      val meetsMemoryRequirements = mem >= calculateTotalMemory(sc)
    +      val meetsCPURequirements = cpus >= (mesosExecutorCores + 
scheduler.CPUS_PER_TASK)
    +      val meetsRequirements =
    +        (meetsMemoryRequirements && meetsCPURequirements) ||
               (slaveIdToExecutorInfo.contains(slaveId) && cpus >= 
scheduler.CPUS_PER_TASK)
    -        val debugstr = if (meetsRequirements) "Accepting" else "Declining"
    -        logDebug(s"$debugstr offer: ${o.getId.getValue} with attributes: "
    -          + s"$offerAttributes mem: $mem cpu: $cpus")
    +      val debugstr = if (meetsRequirements) "Accepting" else "Declining"
    +      logDebug(s"$debugstr offer: ${o.getId.getValue} with attributes: "
    +                           + s"$offerAttributes mem: $mem cpu: $cpus")
    +
    +      meetsRequirements
    +    }
    +
    +    // Decline offers we ruled out immediately
    +    unUsableOffers.foreach(o => d.declineOffer(o.getId))
    +
    +    var availableCores = Math.max(0, maxCores - totalCoresAcquired)
     
    -        meetsRequirements
    +    val workerOffers = (for (o <- usableOffers) yield {
    +      val coresInOffer = getResource(o.getResourcesList, "cpus").toInt
    +      val extraCores = if 
(slaveIdToExecutorInfo.contains(o.getSlaveId.getValue)) {
    +        0D
    +      } else {
    +        // If the Mesos executor has not been started on this slave yet, 
set aside a few
    +        // cores for the Mesos executor by offering fewer cores to the 
Spark executor
    +        mesosExecutorCores
           }
     
    -      // Decline offers we ruled out immediately
    -      unUsableOffers.foreach(o => d.declineOffer(o.getId))
    +      // the cores we can offer for tasks on workers should not exceed 
neither availableCores
    +      // nor cores in the current offer, after accounting for non-task 
cores
    +      val taskCores = Math.min(availableCores - extraCores, coresInOffer - 
extraCores)
     
    -      val workerOffers = usableOffers.map { o =>
    -        val cpus = if 
(slaveIdToExecutorInfo.contains(o.getSlaveId.getValue)) {
    -          getResource(o.getResourcesList, "cpus").toInt
    -        } else {
    -          // If the Mesos executor has not been started on this slave yet, 
set aside a few
    -          // cores for the Mesos executor by offering fewer cores to the 
Spark executor
    -          (getResource(o.getResourcesList, "cpus") - 
mesosExecutorCores).toInt
    -        }
    -        new WorkerOffer(
    +      if (taskCores > 0) {
    +        availableCores -= taskCores + extraCores
    +        Option(new WorkerOffer(
               o.getSlaveId.getValue,
               o.getHostname,
    -          cpus)
    +          taskCores.toInt))
    +      } else {
    +        None
           }
    +    }).flatten
    +
    +    (usableOffers, workerOffers)
    --- End diff --
    
    What happens if we hit the limit and don't run any more tasks on the 
remaining usableOffers? We need to decline those too right?


---
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

Reply via email to