YARN-5077. Fix FSLeafQueue#getFairShare() for queues with zero fairshare. (Yufei Gu via kasha)

(cherry picked from commit 20f2799938c5cad59f59617cdb0f86e137c3c2c0)

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSLeafQueue.java

@@ -43,7 +43,6 @@ import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ActiveUsersManager;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerAppUtils;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.SchedulerApplicationAttempt;
-import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.policies.DominantResourceFairnessPolicy;
 import org.apache.hadoop.yarn.util.resource.Resources;
 
 @Private
@@ -482,8 +481,7 @@ public class FSLeafQueue extends FSQueue {
 
   /**
    * Check whether this queue can run this application master under the
-   * maxAMShare limit. For FIFO and FAIR policies, check if the VCore usage
-   * takes up the entire cluster or maxResources for the queue.
+   * maxAMShare limit.
    * @param amResource
    * @return true if this queue can run
    */
@@ -493,24 +491,25 @@ public class FSLeafQueue extends FSQueue {
     if (Math.abs(maxAMShare - -1.0f) < 0.0001) {
       return true;
     }
-    Resource maxAMResource = Resources.multiply(getFairShare(), maxAMShare);
-    Resource ifRunAMResource = Resources.add(amResourceUsage, amResource);
 
-    boolean overMaxAMShareLimit = policy
-            .checkIfAMResourceUsageOverLimit(ifRunAMResource, maxAMResource);
+    // If FairShare is zero, use min(maxShare, available resource) to compute
+    // maxAMResource
+    Resource maxResource = Resources.clone(getFairShare());
+    if (maxResource.getMemorySize() == 0) {
+      maxResource.setMemory(
+          Math.min(scheduler.getRootQueueMetrics().getAvailableMB(),
+                   getMaxShare().getMemorySize()));
+    }
 
-    // For fair policy and fifo policy which doesn't check VCore usages,
-    // additionally check if the AM takes all available VCores or
-    // over maxResource to avoid deadlock.
-    if (!overMaxAMShareLimit && !policy.equals(
-        SchedulingPolicy.getInstance(DominantResourceFairnessPolicy.class))) {
-      overMaxAMShareLimit =
-         isVCoresOverMaxResource(ifRunAMResource.getVirtualCores()) ||
-         ifRunAMResource.getVirtualCores() >=
-         scheduler.getRootQueueMetrics().getAvailableVirtualCores();
+    if (maxResource.getVirtualCoresSize() == 0) {
+      maxResource.setVirtualCores(Math.min(
+          scheduler.getRootQueueMetrics().getAvailableVirtualCores(),
+          getMaxShare().getVirtualCoresSize()));
     }
 
-    return !overMaxAMShareLimit;
+    Resource maxAMResource = Resources.multiply(maxResource, maxAMShare);
+    Resource ifRunAMResource = Resources.add(amResourceUsage, amResource);
+    return Resources.fitsIn(ifRunAMResource, maxAMResource);
   }
 
   public void addAMResourceUsage(Resource amResource) {

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSQueue.java

@@ -310,25 +310,6 @@ public abstract class FSQueue implements Queue, Schedulable {
     return true;
   }
 
-  /**
-   * Helper method to check if requested VCores are over maxResource.
-   * @param requestedVCores the number of VCores requested
-   * @return true if the number of VCores requested is over the maxResource;
-   *         false otherwise
-   */
-  protected boolean isVCoresOverMaxResource(int requestedVCores) {
-    if (requestedVCores >= scheduler.getAllocationConfiguration().
-        getMaxResources(getName()).getVirtualCores()) {
-      return true;
-    }
-
-    if (getParent() == null) {
-      return false;
-    }
-
-    return getParent().isVCoresOverMaxResource(requestedVCores);
-  }
-
   /**
    * Returns true if queue has at least one app running.
    */

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/SchedulingPolicy.java

@@ -177,17 +177,6 @@ public abstract class SchedulingPolicy {
   public abstract boolean checkIfUsageOverFairShare(
       Resource usage, Resource fairShare);
 
-  /**
-   * Check if a leaf queue's AM resource usage over its limit under this policy
-   *
-   * @param usage {@link Resource} the resource used by application masters
-   * @param maxAMResource {@link Resource} the maximum allowed resource for
-   *                                      application masters
-   * @return true if AM resource usage is over the limit
-   */
-  public abstract boolean checkIfAMResourceUsageOverLimit(
-      Resource usage, Resource maxAMResource);
-
   /**
    * Get headroom by calculating the min of <code>clusterAvailable</code> and
    * (<code>queueFairShare</code> - <code>queueUsage</code>) resources that are

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/policies/DominantResourceFairnessPolicy.java

@@ -93,11 +93,6 @@ public class DominantResourceFairnessPolicy extends SchedulingPolicy {
     return !Resources.fitsIn(usage, fairShare);
   }
 
-  @Override
-  public boolean checkIfAMResourceUsageOverLimit(Resource usage, Resource maxAMResource) {
-    return !Resources.fitsIn(usage, maxAMResource);
-  }
-
   @Override
   public Resource getHeadroom(Resource queueFairShare, Resource queueUsage,
                               Resource maxAvailable) {

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/policies/FairSharePolicy.java

@@ -150,11 +150,6 @@ public class FairSharePolicy extends SchedulingPolicy {
     return Resources.greaterThan(RESOURCE_CALCULATOR, null, usage, fairShare);
   }
 
-  @Override
-  public boolean checkIfAMResourceUsageOverLimit(Resource usage, Resource maxAMResource) {
-    return usage.getMemorySize() > maxAMResource.getMemorySize();
-  }
-
   @Override
   public byte getApplicableDepth() {
     return SchedulingPolicy.DEPTH_ANY;

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/policies/FifoPolicy.java

@@ -113,11 +113,6 @@ public class FifoPolicy extends SchedulingPolicy {
             "as FifoPolicy only works for FSLeafQueue.");
   }
 
-  @Override
-  public boolean checkIfAMResourceUsageOverLimit(Resource usage, Resource maxAMResource) {
-    return usage.getMemorySize() > maxAMResource.getMemorySize();
-  }
-
   @Override
   public Resource getHeadroom(Resource queueFairShare,
                               Resource queueUsage, Resource maxAvailable) {

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/TestFairScheduler.java

@@ -1081,6 +1081,84 @@ public class TestFairScheduler extends FairSchedulerTestBase {
         getCurrentReservation().getMemorySize());
   }
 
+  /**
+   * The test verifies that zero-FairShare queues (because of zero/tiny
+   * weight) can get resources for the AM.
+   */
+  @Test
+  public void testRequestAMResourceInZeroFairShareQueue() throws Exception {
+    conf.set(FairSchedulerConfiguration.ALLOCATION_FILE, ALLOC_FILE);
+
+    PrintWriter out = new PrintWriter(new FileWriter(ALLOC_FILE));
+    out.println("<?xml version=\"1.0\"?>");
+    out.println("<allocations>");
+    out.println("<queue name=\"queue1\">");
+    out.println("<weight>0.0</weight>");
+    out.println("<maxResources>4096mb,10vcores</maxResources>");
+    out.println("<maxAMShare>0.5</maxAMShare>");
+    out.println("</queue>");
+    out.println("<queue name=\"queue2\">");
+    out.println("<weight>2.0</weight>");
+    out.println("</queue>");
+    out.println("<queue name=\"queue3\">");
+    out.println("<weight>0.000001</weight>");
+    out.println("</queue>");
+    out.println("</allocations>");
+    out.close();
+
+    scheduler.init(conf);
+    scheduler.start();
+    scheduler.reinitialize(conf, resourceManager.getRMContext());
+
+    RMNode node =
+            MockNodes.newNodeInfo(1, Resources.createResource(8192, 20),
+                    0, "127.0.0.1");
+    NodeAddedSchedulerEvent nodeEvent = new NodeAddedSchedulerEvent(node);
+    NodeUpdateSchedulerEvent updateEvent = new NodeUpdateSchedulerEvent(node);
+    scheduler.handle(nodeEvent);
+    scheduler.update();
+
+    //create request for non-zero weight queue
+    createSchedulingRequest(1024, "root.queue2", "user2", 1);
+    scheduler.update();
+    scheduler.handle(updateEvent);
+
+    // A managed AM which need 3G memory will not get resource,
+    // since it request more than the maxAMShare (4G * 0.5 = 2G).
+    Resource amResource = Resource.newInstance(1024, 1);
+    int amPriority = RMAppAttemptImpl.AM_CONTAINER_PRIORITY.getPriority();
+    ApplicationAttemptId attId1 = createAppAttemptId(1, 1);
+    createApplicationWithAMResource(attId1, "root.queue1", "user1", amResource);
+    createSchedulingRequestExistingApplication(3 * 1024, 1, amPriority, attId1);
+    FSAppAttempt app1 = scheduler.getSchedulerApp(attId1);
+    scheduler.update();
+    scheduler.handle(updateEvent);
+    assertEquals("Application 1 should not be running",
+            0, app1.getLiveContainers().size());
+
+    // A managed AM which need 2G memory will get resource,
+    // since it request no more than the maxAMShare (4G * 0.5 = 2G).
+    ApplicationAttemptId attId2 = createAppAttemptId(2, 1);
+    createApplicationWithAMResource(attId2, "root.queue1", "user1", amResource);
+    createSchedulingRequestExistingApplication(2 * 1024, 1, amPriority, attId2);
+    FSAppAttempt app2 = scheduler.getSchedulerApp(attId2);
+    scheduler.update();
+    scheduler.handle(updateEvent);
+    assertEquals("Application 2 should be running",
+            1, app2.getLiveContainers().size());
+
+    // A managed AM which need 1G memory will get resource, even thought its
+    // fair share is 0 because its weight is tiny(0.000001).
+    ApplicationAttemptId attId3 = createAppAttemptId(3, 1);
+    createApplicationWithAMResource(attId3, "root.queue3", "user1", amResource);
+    createSchedulingRequestExistingApplication(1024, 1, amPriority, attId3);
+    FSAppAttempt app3 = scheduler.getSchedulerApp(attId3);
+    scheduler.update();
+    scheduler.handle(updateEvent);
+    assertEquals("Application 3 should be running",
+            1, app3.getLiveContainers().size());
+  }
+
     @Test (timeout = 500000)
   public void testContainerReservationNotExceedingQueueMax() throws Exception {
     conf.set(FairSchedulerConfiguration.ALLOCATION_FILE, ALLOC_FILE);