MAPREDUCE-5279. Made MR headroom calculation honor cpu dimension when YARN scheduler resource type is memory plus cpu. Contributed by Peng Zhang and Varun Vasudev.

2014-09-22 09:28:47 -07:00 · 2014-09-22 09:28:47 -07:00 · 376233cdd4
parent 26cba7f35f
commit 376233cdd4
6 changed files with 287 additions and 137 deletions
--- a/hadoop-mapreduce-project/CHANGES.txt
+++ b/hadoop-mapreduce-project/CHANGES.txt
@ -254,6 +254,10 @@ Release 2.6.0 - UNRELEASED
    MAPREDUCE-5891. Improved shuffle error handling across NM restarts
    (Junping Du via jlowe)

+    MAPREDUCE-5279. Made MR headroom calculation honor cpu dimension when YARN
+    scheduler resource type is memory plus cpu. (Peng Zhang and Varun Vasudev
+    via zjshen)
+
  OPTIMIZATIONS

  BUG FIXES
--- a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/RMCommunicator.java
+++ b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/RMCommunicator.java
@ -21,6 +21,7 @@ package org.apache.hadoop.mapreduce.v2.app.rm;
 import java.io.IOException;
 import java.net.InetSocketAddress;
 import java.nio.ByteBuffer;
+import java.util.EnumSet;
 import java.util.Map;
 import java.util.concurrent.ConcurrentLinkedQueue;
 import java.util.concurrent.atomic.AtomicBoolean;
@ -59,6 +60,7 @@ import org.apache.hadoop.yarn.factories.RecordFactory;
 import org.apache.hadoop.yarn.factory.providers.RecordFactoryProvider;

 import com.google.common.annotations.VisibleForTesting;
+import org.apache.hadoop.yarn.proto.YarnServiceProtos.SchedulerResourceTypes;

 /**
 * Registers/unregisters to RM and sends heartbeats to RM.
@ -90,6 +92,8 @@ public abstract class RMCommunicator extends AbstractService
  private volatile boolean shouldUnregister = true;
  private boolean isApplicationMasterRegistered = false;

+  private EnumSet<SchedulerResourceTypes> schedulerResourceTypes;
+
  public RMCommunicator(ClientService clientService, AppContext context) {
    super("RMCommunicator");
    this.clientService = clientService;
@ -98,6 +102,7 @@ public abstract class RMCommunicator extends AbstractService
    this.applicationId = context.getApplicationID();
    this.stopped = new AtomicBoolean(false);
    this.heartbeatCallbacks = new ConcurrentLinkedQueue<Runnable>();
+    this.schedulerResourceTypes = EnumSet.of(SchedulerResourceTypes.MEMORY);
  }

  @Override
@ -163,10 +168,11 @@ public abstract class RMCommunicator extends AbstractService
        setClientToAMToken(response.getClientToAMTokenMasterKey());        
      }
      this.applicationACLs = response.getApplicationACLs();
-      LOG.info("maxContainerCapability: " + maxContainerCapability.getMemory());
+      LOG.info("maxContainerCapability: " + maxContainerCapability);
      String queue = response.getQueue();
      LOG.info("queue: " + queue);
      job.setQueueName(queue);
+      this.schedulerResourceTypes.addAll(response.getSchedulerResourceTypes());
    } catch (Exception are) {
      LOG.error("Exception while registering", are);
      throw new YarnRuntimeException(are);
@ -343,4 +349,8 @@ public abstract class RMCommunicator extends AbstractService
  protected boolean isApplicationMasterRegistered() {
    return isApplicationMasterRegistered;
  }
+
+  public EnumSet<SchedulerResourceTypes> getSchedulerResourceTypes() {
+    return schedulerResourceTypes;
+  }
 }
--- a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/RMContainerAllocator.java
+++ b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/RMContainerAllocator.java
@ -73,6 +73,7 @@ import org.apache.hadoop.yarn.api.records.NodeReport;
 import org.apache.hadoop.yarn.api.records.NodeState;
 import org.apache.hadoop.yarn.api.records.PreemptionMessage;
 import org.apache.hadoop.yarn.api.records.Priority;
+import org.apache.hadoop.yarn.api.records.Resource;
 import org.apache.hadoop.yarn.api.records.Token;
 import org.apache.hadoop.yarn.client.api.NMTokenCache;
 import org.apache.hadoop.yarn.exceptions.YarnRuntimeException;
@ -80,6 +81,7 @@ import org.apache.hadoop.yarn.factory.providers.RecordFactoryProvider;
 import org.apache.hadoop.yarn.security.AMRMTokenIdentifier;
 import org.apache.hadoop.yarn.util.Clock;
 import org.apache.hadoop.yarn.util.RackResolver;
+import org.apache.hadoop.yarn.util.resource.Resources;

 import com.google.common.annotations.VisibleForTesting;

@ -149,8 +151,8 @@ public class RMContainerAllocator extends RMContainerRequestor
  private int lastCompletedTasks = 0;
  
  private boolean recalculateReduceSchedule = false;
-  private int mapResourceRequest;//memory
-  private int reduceResourceRequest;//memory
+  private Resource mapResourceRequest = Resources.none();
+  private Resource reduceResourceRequest = Resources.none();
  
  private boolean reduceStarted = false;
  private float maxReduceRampupLimit = 0;
@ -328,49 +330,61 @@ public class RMContainerAllocator extends RMContainerRequestor
    if (event.getType() == ContainerAllocator.EventType.CONTAINER_REQ) {
      ContainerRequestEvent reqEvent = (ContainerRequestEvent) event;
      JobId jobId = getJob().getID();
-      int supportedMaxContainerCapability =
-          getMaxContainerCapability().getMemory();
+      Resource supportedMaxContainerCapability = getMaxContainerCapability();
      if (reqEvent.getAttemptID().getTaskId().getTaskType().equals(TaskType.MAP)) {
-        if (mapResourceRequest == 0) {
-          mapResourceRequest = reqEvent.getCapability().getMemory();
-          eventHandler.handle(new JobHistoryEvent(jobId, 
-              new NormalizedResourceEvent(org.apache.hadoop.mapreduce.TaskType.MAP,
-                  mapResourceRequest)));
-          LOG.info("mapResourceRequest:"+ mapResourceRequest);
-          if (mapResourceRequest > supportedMaxContainerCapability) {
-            String diagMsg = "MAP capability required is more than the supported " +
-            "max container capability in the cluster. Killing the Job. mapResourceRequest: " +
-                mapResourceRequest + " maxContainerCapability:" + supportedMaxContainerCapability;
+        if (mapResourceRequest.equals(Resources.none())) {
+          mapResourceRequest = reqEvent.getCapability();
+          eventHandler.handle(new JobHistoryEvent(jobId,
+            new NormalizedResourceEvent(
+              org.apache.hadoop.mapreduce.TaskType.MAP, mapResourceRequest
+                .getMemory())));
+          LOG.info("mapResourceRequest:" + mapResourceRequest);
+          if (mapResourceRequest.getMemory() > supportedMaxContainerCapability
+            .getMemory()
+              || mapResourceRequest.getVirtualCores() > supportedMaxContainerCapability
+                .getVirtualCores()) {
+            String diagMsg =
+                "MAP capability required is more than the supported "
+                    + "max container capability in the cluster. Killing the Job. mapResourceRequest: "
+                    + mapResourceRequest + " maxContainerCapability:"
+                    + supportedMaxContainerCapability;
            LOG.info(diagMsg);
-            eventHandler.handle(new JobDiagnosticsUpdateEvent(
-                jobId, diagMsg));
+            eventHandler.handle(new JobDiagnosticsUpdateEvent(jobId, diagMsg));
            eventHandler.handle(new JobEvent(jobId, JobEventType.JOB_KILL));
          }
        }
-        //set the rounded off memory
-        reqEvent.getCapability().setMemory(mapResourceRequest);
+        // set the resources
+        reqEvent.getCapability().setMemory(mapResourceRequest.getMemory());
+        reqEvent.getCapability().setVirtualCores(
+          mapResourceRequest.getVirtualCores());
        scheduledRequests.addMap(reqEvent);//maps are immediately scheduled
      } else {
-        if (reduceResourceRequest == 0) {
-          reduceResourceRequest = reqEvent.getCapability().getMemory();
-          eventHandler.handle(new JobHistoryEvent(jobId, 
-              new NormalizedResourceEvent(
-                  org.apache.hadoop.mapreduce.TaskType.REDUCE,
-                  reduceResourceRequest)));
-          LOG.info("reduceResourceRequest:"+ reduceResourceRequest);
-          if (reduceResourceRequest > supportedMaxContainerCapability) {
-            String diagMsg = "REDUCE capability required is more than the " +
-            		"supported max container capability in the cluster. Killing the " +
-            		"Job. reduceResourceRequest: " + reduceResourceRequest +
-            		" maxContainerCapability:" + supportedMaxContainerCapability;
+        if (reduceResourceRequest.equals(Resources.none())) {
+          reduceResourceRequest = reqEvent.getCapability();
+          eventHandler.handle(new JobHistoryEvent(jobId,
+            new NormalizedResourceEvent(
+              org.apache.hadoop.mapreduce.TaskType.REDUCE,
+              reduceResourceRequest.getMemory())));
+          LOG.info("reduceResourceRequest:" + reduceResourceRequest);
+          if (reduceResourceRequest.getMemory() > supportedMaxContainerCapability
+            .getMemory()
+              || reduceResourceRequest.getVirtualCores() > supportedMaxContainerCapability
+                .getVirtualCores()) {
+            String diagMsg =
+                "REDUCE capability required is more than the "
+                    + "supported max container capability in the cluster. Killing the "
+                    + "Job. reduceResourceRequest: " + reduceResourceRequest
+                    + " maxContainerCapability:"
+                    + supportedMaxContainerCapability;
            LOG.info(diagMsg);
-            eventHandler.handle(new JobDiagnosticsUpdateEvent(
-                jobId, diagMsg));
+            eventHandler.handle(new JobDiagnosticsUpdateEvent(jobId, diagMsg));
            eventHandler.handle(new JobEvent(jobId, JobEventType.JOB_KILL));
          }
        }
-        //set the rounded off memory
-        reqEvent.getCapability().setMemory(reduceResourceRequest);
+        // set the resources
+        reqEvent.getCapability().setMemory(reduceResourceRequest.getMemory());
+        reqEvent.getCapability().setVirtualCores(
+          reduceResourceRequest.getVirtualCores());
        if (reqEvent.getEarlierAttemptFailed()) {
          //add to the front of queue for fail fast
          pendingReduces.addFirst(new ContainerRequest(reqEvent, PRIORITY_REDUCE));
@ -425,34 +439,40 @@ public class RMContainerAllocator extends RMContainerRequestor

  @Private
  @VisibleForTesting
-  synchronized void setReduceResourceRequest(int mem) {
-    this.reduceResourceRequest = mem;
+  synchronized void setReduceResourceRequest(Resource res) {
+    this.reduceResourceRequest = res;
  }

  @Private
  @VisibleForTesting
-  synchronized void setMapResourceRequest(int mem) {
-    this.mapResourceRequest = mem;
+  synchronized void setMapResourceRequest(Resource res) {
+    this.mapResourceRequest = res;
  }

  @Private
  @VisibleForTesting
  void preemptReducesIfNeeded() {
-    if (reduceResourceRequest == 0) {
-      return; //no reduces
+    if (reduceResourceRequest.equals(Resources.none())) {
+      return; // no reduces
    }
    //check if reduces have taken over the whole cluster and there are 
    //unassigned maps
    if (scheduledRequests.maps.size() > 0) {
-      int memLimit = getMemLimit();
-      int availableMemForMap = memLimit - ((assignedRequests.reduces.size() -
-          assignedRequests.preemptionWaitingReduces.size()) * reduceResourceRequest);
-      //availableMemForMap must be sufficient to run atleast 1 map
-      if (availableMemForMap < mapResourceRequest) {
-        //to make sure new containers are given to maps and not reduces
-        //ramp down all scheduled reduces if any
-        //(since reduces are scheduled at higher priority than maps)
-        LOG.info("Ramping down all scheduled reduces:" + scheduledRequests.reduces.size());
+      Resource resourceLimit = getResourceLimit();
+      Resource availableResourceForMap =
+          Resources.subtract(
+            resourceLimit,
+            Resources.multiply(reduceResourceRequest,
+              assignedRequests.reduces.size()
+                  - assignedRequests.preemptionWaitingReduces.size()));
+      // availableMemForMap must be sufficient to run at least 1 map
+      if (ResourceCalculatorUtils.computeAvailableContainers(availableResourceForMap,
+        mapResourceRequest, getSchedulerResourceTypes()) <= 0) {
+        // to make sure new containers are given to maps and not reduces
+        // ramp down all scheduled reduces if any
+        // (since reduces are scheduled at higher priority than maps)
+        LOG.info("Ramping down all scheduled reduces:"
+            + scheduledRequests.reduces.size());
        for (ContainerRequest req : scheduledRequests.reduces.values()) {
          pendingReduces.add(req);
        }
@ -462,17 +482,25 @@ public class RMContainerAllocator extends RMContainerRequestor
        //hanging around for a while
        int hangingMapRequests = getNumOfHangingRequests(scheduledRequests.maps);
        if (hangingMapRequests > 0) {
-          //preempt for making space for at least one map
-          int premeptionLimit = Math.max(mapResourceRequest,
-              (int) (maxReducePreemptionLimit * memLimit));
+          // preempt for making space for at least one map
+          int preemptionReduceNumForOneMap =
+              ResourceCalculatorUtils.divideAndCeilContainers(mapResourceRequest,
+                reduceResourceRequest, getSchedulerResourceTypes());
+          int preemptionReduceNumForPreemptionLimit =
+              ResourceCalculatorUtils.divideAndCeilContainers(
+                Resources.multiply(resourceLimit, maxReducePreemptionLimit),
+                reduceResourceRequest, getSchedulerResourceTypes());
+          int preemptionReduceNumForAllMaps =
+              ResourceCalculatorUtils.divideAndCeilContainers(
+                Resources.multiply(mapResourceRequest, hangingMapRequests),
+                reduceResourceRequest, getSchedulerResourceTypes());
+          int toPreempt =
+              Math.min(Math.max(preemptionReduceNumForOneMap,
+                preemptionReduceNumForPreemptionLimit),
+                preemptionReduceNumForAllMaps);

-          int preemptMem = Math.min(hangingMapRequests * mapResourceRequest,
-              premeptionLimit);
-
-          int toPreempt = (int) Math.ceil((float) preemptMem / reduceResourceRequest);
-          toPreempt = Math.min(toPreempt, assignedRequests.reduces.size());
-
-          LOG.info("Going to preempt " + toPreempt + " due to lack of space for maps");
+          LOG.info("Going to preempt " + toPreempt
+              + " due to lack of space for maps");
          assignedRequests.preemptReduce(toPreempt);
        }
      }
@ -497,7 +525,7 @@ public class RMContainerAllocator extends RMContainerRequestor
      int totalMaps, int completedMaps,
      int scheduledMaps, int scheduledReduces,
      int assignedMaps, int assignedReduces,
-      int mapResourceReqt, int reduceResourceReqt,
+      Resource mapResourceReqt, Resource reduceResourceReqt,
      int numPendingReduces,
      float maxReduceRampupLimit, float reduceSlowStart) {
    
@ -505,8 +533,12 @@ public class RMContainerAllocator extends RMContainerRequestor
      return;
    }
    
-    int headRoom = getAvailableResources() != null ?
-        getAvailableResources().getMemory() : 0;
+    // get available resources for this job
+    Resource headRoom = getAvailableResources();
+    if (headRoom == null) {
+      headRoom = Resources.none();
+    }
+
    LOG.info("Recalculating schedule, headroom=" + headRoom);
    
    //check for slow start
@ -540,49 +572,60 @@ public class RMContainerAllocator extends RMContainerRequestor
      completedMapPercent = 1;
    }
    
-    int netScheduledMapMem = 
-        (scheduledMaps + assignedMaps) * mapResourceReqt;
+    Resource netScheduledMapResource =
+        Resources.multiply(mapResourceReqt, (scheduledMaps + assignedMaps));

-    int netScheduledReduceMem = 
-        (scheduledReduces + assignedReduces) * reduceResourceReqt;
+    Resource netScheduledReduceResource =
+        Resources.multiply(reduceResourceReqt,
+          (scheduledReduces + assignedReduces));
+
+    Resource finalMapResourceLimit;
+    Resource finalReduceResourceLimit;

-    int finalMapMemLimit = 0;
-    int finalReduceMemLimit = 0;
-    
    // ramp up the reduces based on completed map percentage
-    int totalMemLimit = getMemLimit();
-    int idealReduceMemLimit = 
-        Math.min(
-            (int)(completedMapPercent * totalMemLimit),
-            (int) (maxReduceRampupLimit * totalMemLimit));
-    int idealMapMemLimit = totalMemLimit - idealReduceMemLimit;
+    Resource totalResourceLimit = getResourceLimit();
+
+    Resource idealReduceResourceLimit =
+        Resources.multiply(totalResourceLimit,
+          Math.min(completedMapPercent, maxReduceRampupLimit));
+    Resource ideaMapResourceLimit =
+        Resources.subtract(totalResourceLimit, idealReduceResourceLimit);

    // check if there aren't enough maps scheduled, give the free map capacity
-    // to reduce
-    if (idealMapMemLimit > netScheduledMapMem) {
-      int unusedMapMemLimit = idealMapMemLimit - netScheduledMapMem;
-      finalReduceMemLimit = idealReduceMemLimit + unusedMapMemLimit;
-      finalMapMemLimit = totalMemLimit - finalReduceMemLimit;
+    // to reduce.
+    // Even when container number equals, there may be unused resources in one
+    // dimension
+    if (ResourceCalculatorUtils.computeAvailableContainers(ideaMapResourceLimit,
+      mapResourceReqt, getSchedulerResourceTypes()) >= (scheduledMaps + assignedMaps)) {
+      // enough resource given to maps, given the remaining to reduces
+      Resource unusedMapResourceLimit =
+          Resources.subtract(ideaMapResourceLimit, netScheduledMapResource);
+      finalReduceResourceLimit =
+          Resources.add(idealReduceResourceLimit, unusedMapResourceLimit);
+      finalMapResourceLimit =
+          Resources.subtract(totalResourceLimit, finalReduceResourceLimit);
    } else {
-      finalMapMemLimit = idealMapMemLimit;
-      finalReduceMemLimit = idealReduceMemLimit;
+      finalMapResourceLimit = ideaMapResourceLimit;
+      finalReduceResourceLimit = idealReduceResourceLimit;
    }
-    
-    LOG.info("completedMapPercent " + completedMapPercent +
-        " totalMemLimit:" + totalMemLimit +
-        " finalMapMemLimit:" + finalMapMemLimit +
-        " finalReduceMemLimit:" + finalReduceMemLimit + 
-        " netScheduledMapMem:" + netScheduledMapMem +
-        " netScheduledReduceMem:" + netScheduledReduceMem);
-    
-    int rampUp = 
-        (finalReduceMemLimit - netScheduledReduceMem) / reduceResourceReqt;
-    
+
+    LOG.info("completedMapPercent " + completedMapPercent
+        + " totalResourceLimit:" + totalResourceLimit
+        + " finalMapResourceLimit:" + finalMapResourceLimit
+        + " finalReduceResourceLimit:" + finalReduceResourceLimit
+        + " netScheduledMapResource:" + netScheduledMapResource
+        + " netScheduledReduceResource:" + netScheduledReduceResource);
+
+    int rampUp =
+        ResourceCalculatorUtils.computeAvailableContainers(Resources.subtract(
+                finalReduceResourceLimit, netScheduledReduceResource),
+            reduceResourceReqt, getSchedulerResourceTypes());
+
    if (rampUp > 0) {
      rampUp = Math.min(rampUp, numPendingReduces);
      LOG.info("Ramping up " + rampUp);
      rampUpReduces(rampUp);
-    } else if (rampUp < 0){
+    } else if (rampUp < 0) {
      int rampDown = -1 * rampUp;
      rampDown = Math.min(rampDown, scheduledReduces);
      LOG.info("Ramping down " + rampDown);
@ -618,8 +661,10 @@ public class RMContainerAllocator extends RMContainerRequestor
  
  @SuppressWarnings("unchecked")
  private List<Container> getResources() throws Exception {
-    int headRoom = getAvailableResources() != null
-        ? getAvailableResources().getMemory() : 0;//first time it would be null
+    // will be null the first time
+    Resource headRoom =
+        getAvailableResources() == null ? Resources.none() :
+            Resources.clone(getAvailableResources());
    AllocateResponse response;
    /*
     * If contact with RM is lost, the AM will wait MR_AM_TO_RM_WAIT_INTERVAL_MS
@ -670,7 +715,9 @@ public class RMContainerAllocator extends RMContainerRequestor
        throw new YarnRuntimeException(msg);
      }
    }
-    int newHeadRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;
+    Resource newHeadRoom =
+        getAvailableResources() == null ? Resources.none()
+            : getAvailableResources();
    List<Container> newContainers = response.getAllocatedContainers();
    // Setting NMTokens
    if (response.getNMTokens() != null) {
@ -694,10 +741,11 @@ public class RMContainerAllocator extends RMContainerRequestor
          new PreemptionContext(assignedRequests), preemptReq);
    }

-    if (newContainers.size() + finishedContainers.size() > 0 || headRoom != newHeadRoom) {
+    if (newContainers.size() + finishedContainers.size() > 0
+        || !headRoom.equals(newHeadRoom)) {
      //something changed
      recalculateReduceSchedule = true;
-      if (LOG.isDebugEnabled() && headRoom != newHeadRoom) {
+      if (LOG.isDebugEnabled() && !headRoom.equals(newHeadRoom)) {
        LOG.debug("headroom=" + newHeadRoom);
      }
    }
@ -802,10 +850,18 @@ public class RMContainerAllocator extends RMContainerRequestor
  }

  @Private
-  public int getMemLimit() {
-    int headRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;
-    return headRoom + assignedRequests.maps.size() * mapResourceRequest +
-       assignedRequests.reduces.size() * reduceResourceRequest;
+  public Resource getResourceLimit() {
+    Resource headRoom = getAvailableResources();
+    if (headRoom == null) {
+      headRoom = Resources.none();
+    }
+    Resource assignedMapResource =
+        Resources.multiply(mapResourceRequest, assignedRequests.maps.size());
+    Resource assignedReduceResource =
+        Resources.multiply(reduceResourceRequest,
+          assignedRequests.reduces.size());
+    return Resources.add(headRoom,
+      Resources.add(assignedMapResource, assignedReduceResource));
  }

  @Private
@ -914,10 +970,11 @@ public class RMContainerAllocator extends RMContainerRequestor
        // a container to be assigned
        boolean isAssignable = true;
        Priority priority = allocated.getPriority();
-        int allocatedMemory = allocated.getResource().getMemory();
+        Resource allocatedResource = allocated.getResource();
        if (PRIORITY_FAST_FAIL_MAP.equals(priority) 
            || PRIORITY_MAP.equals(priority)) {
-          if (allocatedMemory < mapResourceRequest
+          if (ResourceCalculatorUtils.computeAvailableContainers(allocatedResource,
+              mapResourceRequest, getSchedulerResourceTypes()) <= 0
              || maps.isEmpty()) {
            LOG.info("Cannot assign container " + allocated 
                + " for a map as either "
@ -928,7 +985,8 @@ public class RMContainerAllocator extends RMContainerRequestor
          }
        } 
        else if (PRIORITY_REDUCE.equals(priority)) {
-          if (allocatedMemory < reduceResourceRequest
+          if (ResourceCalculatorUtils.computeAvailableContainers(allocatedResource,
+              reduceResourceRequest, getSchedulerResourceTypes()) <= 0
              || reduces.isEmpty()) {
            LOG.info("Cannot assign container " + allocated 
                + " for a reduce as either "
--- a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/ResourceCalculatorUtils.java
+++ b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/rm/ResourceCalculatorUtils.java
@ -0,0 +1,52 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hadoop.mapreduce.v2.app.rm;
+
+import org.apache.hadoop.yarn.api.records.Resource;
+import org.apache.hadoop.yarn.proto.YarnServiceProtos.SchedulerResourceTypes;
+import org.apache.hadoop.yarn.util.Records;
+
+import java.util.EnumSet;
+
+public class ResourceCalculatorUtils {
+  public static int divideAndCeil(int a, int b) {
+    if (b == 0) {
+      return 0;
+    }
+    return (a + (b - 1)) / b;
+  }
+
+  public static int computeAvailableContainers(Resource available,
+      Resource required, EnumSet<SchedulerResourceTypes> resourceTypes) {
+    if (resourceTypes.contains(SchedulerResourceTypes.CPU)) {
+      return Math.min(available.getMemory() / required.getMemory(),
+        available.getVirtualCores() / required.getVirtualCores());
+    }
+    return available.getMemory() / required.getMemory();
+  }
+
+  public static int divideAndCeilContainers(Resource required, Resource factor,
+      EnumSet<SchedulerResourceTypes> resourceTypes) {
+    if (resourceTypes.contains(SchedulerResourceTypes.CPU)) {
+      return Math.max(divideAndCeil(required.getMemory(), factor.getMemory()),
+        divideAndCeil(required.getVirtualCores(), factor.getVirtualCores()));
+    }
+    return divideAndCeil(required.getMemory(), factor.getMemory());
+  }
+}
--- a/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/test/java/org/apache/hadoop/mapreduce/v2/app/rm/TestRMContainerAllocator.java
+++ b/hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/test/java/org/apache/hadoop/mapreduce/v2/app/rm/TestRMContainerAllocator.java
@ -18,6 +18,7 @@

 package org.apache.hadoop.mapreduce.v2.app.rm;

+import static org.mockito.Matchers.any;
 import static org.mockito.Matchers.anyFloat;
 import static org.mockito.Matchers.anyInt;
 import static org.mockito.Matchers.isA;
@ -30,19 +31,14 @@ import static org.mockito.Mockito.verify;
 import static org.mockito.Mockito.when;

 import java.io.IOException;
-import java.util.ArrayList;
-import java.util.Arrays;
-import java.util.HashMap;
-import java.util.HashSet;
-import java.util.Iterator;
-import java.util.List;
-import java.util.Map;
-import java.util.Set;
+import java.util.*;
 import java.util.concurrent.atomic.AtomicBoolean;

 import org.apache.hadoop.mapreduce.v2.app.AppContext;
 import org.apache.hadoop.mapreduce.v2.app.ClusterInfo;
 import org.apache.hadoop.mapreduce.v2.app.MRApp;
+import org.apache.hadoop.yarn.proto.YarnServiceProtos.SchedulerResourceTypes;
+import org.apache.hadoop.yarn.server.utils.BuilderUtils;
 import org.junit.Assert;
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
@ -458,8 +454,8 @@ public class TestRMContainerAllocator {
            0, 0, 0, 0, 0, 0, "jobfile", null, false, ""));
    MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
        appAttemptId, mockJob, new SystemClock());
-    allocator.setMapResourceRequest(1024);
-    allocator.setReduceResourceRequest(1024);
+    allocator.setMapResourceRequest(BuilderUtils.newResource(1024, 1));
+    allocator.setReduceResourceRequest(BuilderUtils.newResource(1024, 1));
    RMContainerAllocator.AssignedRequests assignedRequests =
        allocator.getAssignedRequests();
    RMContainerAllocator.ScheduledRequests scheduledRequests =
@ -478,7 +474,7 @@ public class TestRMContainerAllocator {

  @Test(timeout = 30000)
  public void testNonAggressivelyPreemptReducers() throws Exception {
-    LOG.info("Running testPreemptReducers");
+    LOG.info("Running testNonAggressivelyPreemptReducers");

    final int preemptThreshold = 2; //sec
    Configuration conf = new Configuration();
@ -513,8 +509,8 @@ public class TestRMContainerAllocator {
    clock.setTime(1);
    MyContainerAllocator allocator = new MyContainerAllocator(rm, conf,
        appAttemptId, mockJob, clock);
-    allocator.setMapResourceRequest(1024);
-    allocator.setReduceResourceRequest(1024);
+    allocator.setMapResourceRequest(BuilderUtils.newResource(1024, 1));
+    allocator.setReduceResourceRequest(BuilderUtils.newResource(1024, 1));
    RMContainerAllocator.AssignedRequests assignedRequests =
        allocator.getAssignedRequests();
    RMContainerAllocator.ScheduledRequests scheduledRequests =
@ -1774,17 +1770,19 @@ public class TestRMContainerAllocator {
    int scheduledReduces = 0;
    int assignedMaps = 2;
    int assignedReduces = 0;
-    int mapResourceReqt = 1024;
-    int reduceResourceReqt = 2*1024;
+    Resource mapResourceReqt = BuilderUtils.newResource(1024, 1);
+    Resource reduceResourceReqt = BuilderUtils.newResource(2 * 1024, 1);
    int numPendingReduces = 4;
    float maxReduceRampupLimit = 0.5f;
    float reduceSlowStart = 0.2f;
    
    RMContainerAllocator allocator = mock(RMContainerAllocator.class);
-    doCallRealMethod().when(allocator).
-        scheduleReduces(anyInt(), anyInt(), anyInt(), anyInt(), anyInt(), 
-            anyInt(), anyInt(), anyInt(), anyInt(), anyFloat(), anyFloat());
-    
+    doCallRealMethod().when(allocator).scheduleReduces(anyInt(), anyInt(),
+        anyInt(), anyInt(), anyInt(), anyInt(), any(Resource.class),
+        any(Resource.class), anyInt(), anyFloat(), anyFloat());
+    doReturn(EnumSet.of(SchedulerResourceTypes.MEMORY)).when(allocator)
+      .getSchedulerResourceTypes();
+
    // Test slow-start
    allocator.scheduleReduces(
        totalMaps, succeededMaps, 
@ -1808,6 +1806,7 @@ public class TestRMContainerAllocator {
    verify(allocator, never()).scheduleAllReduces();

    succeededMaps = 3;
+    doReturn(BuilderUtils.newResource(0, 0)).when(allocator).getResourceLimit();
    allocator.scheduleReduces(
        totalMaps, succeededMaps, 
        scheduledMaps, scheduledReduces, 
@ -1818,7 +1817,8 @@ public class TestRMContainerAllocator {
    verify(allocator, times(1)).setIsReduceStarted(true);
    
    // Test reduce ramp-up
-    doReturn(100 * 1024).when(allocator).getMemLimit();
+    doReturn(BuilderUtils.newResource(100 * 1024, 100 * 1)).when(allocator)
+      .getResourceLimit();
    allocator.scheduleReduces(
        totalMaps, succeededMaps, 
        scheduledMaps, scheduledReduces, 
@ -1831,13 +1831,14 @@ public class TestRMContainerAllocator {

    // Test reduce ramp-down
    scheduledReduces = 3;
-    doReturn(10 * 1024).when(allocator).getMemLimit();
+    doReturn(BuilderUtils.newResource(10 * 1024, 10 * 1)).when(allocator)
+      .getResourceLimit();
    allocator.scheduleReduces(
-        totalMaps, succeededMaps, 
-        scheduledMaps, scheduledReduces, 
-        assignedMaps, assignedReduces, 
-        mapResourceReqt, reduceResourceReqt, 
-        numPendingReduces, 
+        totalMaps, succeededMaps,
+        scheduledMaps, scheduledReduces,
+        assignedMaps, assignedReduces,
+        mapResourceReqt, reduceResourceReqt,
+        numPendingReduces,
        maxReduceRampupLimit, reduceSlowStart);
    verify(allocator).rampDownReduces(anyInt());

@ -1846,7 +1847,8 @@ public class TestRMContainerAllocator {
    // should be invoked twice.
    scheduledMaps = 2;
    assignedReduces = 2;
-    doReturn(10 * 1024).when(allocator).getMemLimit();
+    doReturn(BuilderUtils.newResource(10 * 1024, 10 * 1)).when(allocator)
+      .getResourceLimit();
    allocator.scheduleReduces(
        totalMaps, succeededMaps, 
        scheduledMaps, scheduledReduces, 
@ -1855,6 +1857,30 @@ public class TestRMContainerAllocator {
        numPendingReduces, 
        maxReduceRampupLimit, reduceSlowStart);
    verify(allocator, times(2)).rampDownReduces(anyInt());
+
+    doReturn(
+        EnumSet.of(SchedulerResourceTypes.MEMORY, SchedulerResourceTypes.CPU))
+        .when(allocator).getSchedulerResourceTypes();
+
+    // Test ramp-down when enough memory but not enough cpu resource
+    scheduledMaps = 10;
+    assignedReduces = 0;
+    doReturn(BuilderUtils.newResource(100 * 1024, 5 * 1)).when(allocator)
+        .getResourceLimit();
+    allocator.scheduleReduces(totalMaps, succeededMaps, scheduledMaps,
+        scheduledReduces, assignedMaps, assignedReduces, mapResourceReqt,
+        reduceResourceReqt, numPendingReduces, maxReduceRampupLimit,
+        reduceSlowStart);
+    verify(allocator, times(3)).rampDownReduces(anyInt());
+
+    // Test ramp-down when enough cpu but not enough memory resource
+    doReturn(BuilderUtils.newResource(10 * 1024, 100 * 1)).when(allocator)
+        .getResourceLimit();
+    allocator.scheduleReduces(totalMaps, succeededMaps, scheduledMaps,
+        scheduledReduces, assignedMaps, assignedReduces, mapResourceReqt,
+        reduceResourceReqt, numPendingReduces, maxReduceRampupLimit,
+        reduceSlowStart);
+    verify(allocator, times(4)).rampDownReduces(anyInt());
  }

  private static class RecalculateContainerAllocator extends MyContainerAllocator {
@ -1868,7 +1894,7 @@ public class TestRMContainerAllocator {
    @Override
    public void scheduleReduces(int totalMaps, int completedMaps,
        int scheduledMaps, int scheduledReduces, int assignedMaps,
-        int assignedReduces, int mapResourceReqt, int reduceResourceReqt,
+        int assignedReduces, Resource mapResourceReqt, Resource reduceResourceReqt,
        int numPendingReduces, float maxReduceRampupLimit, float reduceSlowStart) {
      recalculatedReduceSchedule = true;
    }
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-common/src/main/java/org/apache/hadoop/yarn/util/resource/Resources.java
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-common/src/main/java/org/apache/hadoop/yarn/util/resource/Resources.java
@ -18,12 +18,12 @@

 package org.apache.hadoop.yarn.util.resource;

-import org.apache.hadoop.classification.InterfaceAudience.Private;
+import org.apache.hadoop.classification.InterfaceAudience;
 import org.apache.hadoop.classification.InterfaceStability.Unstable;
 import org.apache.hadoop.yarn.api.records.Resource;
 import org.apache.hadoop.yarn.util.Records;

-@Private
+@InterfaceAudience.LimitedPrivate({"YARN", "MapReduce"})
@Unstable
 public class Resources {