YARN-10688. ClusterMetrics should support GPU capacity related metrics.. Contributed by Qi Zhu.

(cherry picked from commit 49f89f1d3d)

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ClusterMetrics.java

@@ -34,6 +34,8 @@ import org.apache.hadoop.metrics2.lib.MutableGaugeLong;
 import org.apache.hadoop.metrics2.lib.MutableRate;
 import org.apache.hadoop.yarn.api.records.Resource;
 import org.apache.hadoop.thirdparty.com.google.common.annotations.VisibleForTesting;
+import org.apache.hadoop.yarn.api.records.ResourceInformation;
+import org.apache.hadoop.yarn.util.resource.ResourceUtils;
 
 @InterfaceAudience.Private
 @Metrics(context="yarn")
@@ -56,6 +58,7 @@ public class ClusterMetrics {
   @Metric("Vcore Utilization") MutableGaugeLong utilizedVirtualCores;
   @Metric("Memory Capability") MutableGaugeLong capabilityMB;
   @Metric("Vcore Capability") MutableGaugeLong capabilityVirtualCores;
+  @Metric("GPU Capability") MutableGaugeLong capabilityGPUs;
 
   private static final MetricsInfo RECORD_INFO = info("ClusterMetrics",
   "Metrics for the Yarn Cluster");
@@ -206,10 +209,24 @@ public class ClusterMetrics {
     return capabilityVirtualCores.value();
   }
 
+  public long getCapabilityGPUs() {
+    if (capabilityGPUs == null) {
+      return 0;
+    }
+
+    return capabilityGPUs.value();
+  }
+
   public void incrCapability(Resource res) {
     if (res != null) {
       capabilityMB.incr(res.getMemorySize());
       capabilityVirtualCores.incr(res.getVirtualCores());
+      Integer gpuIndex = ResourceUtils.getResourceTypeIndex()
+          .get(ResourceInformation.GPU_URI);
+      if (gpuIndex != null) {
+        capabilityGPUs.incr(res.
+            getResourceValue(ResourceInformation.GPU_URI));
+      }
     }
   }
 
@@ -217,6 +234,12 @@ public class ClusterMetrics {
     if (res != null) {
       capabilityMB.decr(res.getMemorySize());
       capabilityVirtualCores.decr(res.getVirtualCores());
+      Integer gpuIndex = ResourceUtils.getResourceTypeIndex()
+          .get(ResourceInformation.GPU_URI);
+      if (gpuIndex != null) {
+        capabilityGPUs.decr(res.
+            getResourceValue(ResourceInformation.GPU_URI));
+      }
     }
   }
 

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCSAllocateCustomResource.java

@@ -22,18 +22,22 @@ import org.apache.commons.io.FileUtils;
 import org.apache.hadoop.yarn.api.records.ContainerId;
 import org.apache.hadoop.yarn.api.records.Resource;
 import org.apache.hadoop.yarn.conf.YarnConfiguration;
+import org.apache.hadoop.yarn.server.resourcemanager.ClusterMetrics;
 import org.apache.hadoop.yarn.server.resourcemanager.MockAM;
 import org.apache.hadoop.yarn.server.resourcemanager.MockNM;
 import org.apache.hadoop.yarn.server.resourcemanager.MockRM;
 import org.apache.hadoop.yarn.server.resourcemanager.MockRMAppSubmissionData;
 import org.apache.hadoop.yarn.server.resourcemanager.MockRMAppSubmitter;
+import org.apache.hadoop.yarn.server.resourcemanager.MockNodes;
 import org.apache.hadoop.yarn.server.resourcemanager.nodelabels.NullRMNodeLabelsManager;
 import org.apache.hadoop.yarn.server.resourcemanager.nodelabels.RMNodeLabelsManager;
 import org.apache.hadoop.yarn.server.resourcemanager.resource.TestResourceProfiles;
 import org.apache.hadoop.yarn.server.resourcemanager.rmapp.RMApp;
 import org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNode;
+import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ClusterNodeTracker;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceScheduler;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerApp;
+import org.apache.hadoop.yarn.server.resourcemanager.scheduler.common.fica.FiCaSchedulerNode;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.NodeUpdateSchedulerEvent;
 import org.apache.hadoop.yarn.util.resource.DominantResourceCalculator;
 import org.apache.hadoop.yarn.util.resource.ResourceCalculator;
@@ -47,8 +51,12 @@ import org.junit.Test;
 import java.io.File;
 import java.io.IOException;
 import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
 
+import static org.apache.hadoop.yarn.api.records.ResourceInformation.GPU_URI;
 import static org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacitySchedulerConfiguration.MAXIMUM_ALLOCATION_MB;
+import static org.junit.Assert.assertEquals;
 
 /**
  * Test case for custom resource container allocation.
@@ -64,6 +72,9 @@ public class TestCSAllocateCustomResource {
 
   private final int g = 1024;
 
+  private ClusterNodeTracker<FiCaSchedulerNode> nodeTracker;
+  private ClusterMetrics metrics;
+
   @Before
   public void setUp() throws Exception {
     conf = new YarnConfiguration();
@@ -182,4 +193,57 @@ public class TestCSAllocateCustomResource {
             .getResourceValue("yarn.io/gpu"));
     rm.close();
   }
+
+  @Test
+  public void testClusterMetricsWithGPU()
+      throws Exception {
+    metrics = ClusterMetrics.getMetrics();
+    // reset resource types
+    ResourceUtils.resetResourceTypes();
+    String resourceTypesFileName = "resource-types-test.xml";
+    File source = new File(
+        conf.getClassLoader().getResource(resourceTypesFileName).getFile());
+    resourceTypesFile = new File(source.getParent(), "resource-types.xml");
+    FileUtils.copyFile(source, resourceTypesFile);
+
+    CapacitySchedulerConfiguration newConf =
+        (CapacitySchedulerConfiguration) TestUtils
+            .getConfigurationWithMultipleQueues(conf);
+    newConf.setClass(CapacitySchedulerConfiguration.RESOURCE_CALCULATOR_CLASS,
+        DominantResourceCalculator.class, ResourceCalculator.class);
+    //start RM
+    MockRM rm = new MockRM(newConf);
+    rm.start();
+
+    nodeTracker = new ClusterNodeTracker<>();
+    MockNodes.resetHostIds();
+    Resource nodeResource = Resource.newInstance(4096, 4,
+        Collections.singletonMap(GPU_URI, 4L));
+    List<RMNode> rmNodes =
+        MockNodes.newNodes(2, 4, nodeResource);
+    for (RMNode rmNode : rmNodes) {
+      nodeTracker.addNode(new FiCaSchedulerNode(rmNode, false));
+    }
+
+    // Check GPU inc related cluster metrics.
+    assertEquals("Cluster Capability Memory incorrect",
+        metrics.getCapabilityMB(), (4096 * 8));
+    assertEquals("Cluster Capability Vcores incorrect",
+        metrics.getCapabilityVirtualCores(), 4 * 8);
+    assertEquals("Cluster Capability GPUs incorrect",
+        metrics.getCapabilityGPUs(), 4 * 8);
+
+    for (RMNode rmNode : rmNodes) {
+      nodeTracker.removeNode(rmNode.getNodeID());
+    }
+
+    // Check GPU dec related cluster metrics.
+    assertEquals("Cluster Capability Memory incorrect",
+        metrics.getCapabilityMB(), 0);
+    assertEquals("Cluster Capability Vcores incorrect",
+        metrics.getCapabilityVirtualCores(), 0);
+    assertEquals("Cluster Capability GPUs incorrect",
+        metrics.getCapabilityGPUs(), 0);
+    ClusterMetrics.destroy();
+  }
 }