You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-commits@hadoop.apache.org by tu...@apache.org on 2013/05/10 00:15:59 UTC

svn commit: r1480809 - in /hadoop/common/trunk/hadoop-yarn-project: ./ hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/ hadoop-yarn/hadoop-yarn-server/hadoop-yarn-s...

Author: tucu
Date: Thu May  9 22:15:58 2013
New Revision: 1480809

URL: http://svn.apache.org/r1480809
Log:
YARN-655. Fair scheduler metrics should subtract allocated memory from available memory. (sandyr via tucu)

Modified:
    hadoop/common/trunk/hadoop-yarn-project/CHANGES.txt
    hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/QueueMetrics.java
    hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java
    hadoop/common/trunk/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

Modified: hadoop/common/trunk/hadoop-yarn-project/CHANGES.txt
URL: http://svn.apache.org/viewvc/hadoop/common/trunk/hadoop-yarn-project/CHANGES.txt?rev=1480809&r1=1480808&r2=1480809&view=diff
==============================================================================
--- hadoop/common/trunk/hadoop-yarn-project/CHANGES.txt (original)
+++ hadoop/common/trunk/hadoop-yarn-project/CHANGES.txt Thu May  9 22:15:58 2013
@@ -367,6 +367,9 @@ Release 2.0.5-beta - UNRELEASED
 
     YARN-637. FS: maxAssign is not honored. (kkambatl via tucu)
 
+    YARN-655. Fair scheduler metrics should subtract allocated memory from 
+    available memory. (sandyr via tucu)
+
 Release 2.0.4-alpha - 2013-04-25 
 
   INCOMPATIBLE CHANGES

Modified: hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/QueueMetrics.java
URL: http://svn.apache.org/viewvc/hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/QueueMetrics.java?rev=1480809&r1=1480808&r2=1480809&view=diff
==============================================================================
--- hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/QueueMetrics.java (original)
+++ hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/QueueMetrics.java Thu May  9 22:15:58 2013
@@ -43,6 +43,7 @@ import org.apache.hadoop.yarn.api.record
 import org.apache.hadoop.yarn.conf.YarnConfiguration;
 import org.apache.hadoop.yarn.server.resourcemanager.resource.Resources;
 import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptState;
+import org.apache.hadoop.yarn.util.BuilderUtils;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
@@ -425,6 +426,10 @@ public class QueueMetrics implements Met
   public int getAppsFailed() {
     return appsFailed.value();
   }
+  
+  public Resource getAllocatedResources() {
+    return BuilderUtils.newResource(allocatedMB.value(), 0);
+  }
 
   public int getAllocatedMB() {
     return allocatedMB.value();

Modified: hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java
URL: http://svn.apache.org/viewvc/hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java?rev=1480809&r1=1480808&r2=1480809&view=diff
==============================================================================
--- hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java (original)
+++ hadoop/common/trunk/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java Thu May  9 22:15:58 2013
@@ -234,10 +234,6 @@ public class FairScheduler implements Re
     // Recursively compute fair shares for all queues
     // and update metrics
     rootQueue.recomputeShares();
-
-    // Update recorded capacity of root queue (child queues are updated
-    // when fair share is calculated).
-    rootMetrics.setAvailableResourcesToQueue(clusterCapacity);
   }
 
   /**
@@ -670,6 +666,7 @@ public class FairScheduler implements Re
     } else {
       application.containerCompleted(rmContainer, containerStatus, event);
       node.releaseContainer(container);
+      updateRootQueueMetrics();
     }
 
     LOG.info("Application " + applicationAttemptId +
@@ -681,6 +678,7 @@ public class FairScheduler implements Re
   private synchronized void addNode(RMNode node) {
     nodes.put(node.getNodeID(), new FSSchedulerNode(node));
     Resources.addTo(clusterCapacity, node.getTotalCapability());
+    updateRootQueueMetrics();
 
     LOG.info("Added node " + node.getNodeAddress() +
         " cluster capacity: " + clusterCapacity);
@@ -689,6 +687,7 @@ public class FairScheduler implements Re
   private synchronized void removeNode(RMNode rmNode) {
     FSSchedulerNode node = nodes.get(rmNode.getNodeID());
     Resources.subtractFrom(clusterCapacity, rmNode.getTotalCapability());
+    updateRootQueueMetrics();
 
     // Remove running containers
     List<RMContainer> runningContainers = node.getRunningContainers();
@@ -901,6 +900,7 @@ public class FairScheduler implements Re
         if ((assignedContainers >= maxAssign) && (maxAssign > 0)) { break; }
       }
     }
+    updateRootQueueMetrics();
   }
 
   @Override
@@ -922,6 +922,18 @@ public class FairScheduler implements Re
     }
     return new SchedulerAppReport(applications.get(appAttemptId));
   }
+  
+  /**
+   * Subqueue metrics might be a little out of date because fair shares are
+   * recalculated at the update interval, but the root queue metrics needs to
+   * be updated synchronously with allocations and completions so that cluster
+   * metrics will be consistent.
+   */
+  private void updateRootQueueMetrics() {
+    rootMetrics.setAvailableResourcesToQueue(
+        Resources.subtract(
+            clusterCapacity, rootMetrics.getAllocatedResources()));
+  }
 
   @Override
   public QueueMetrics getRootQueueMetrics() {

Modified: hadoop/common/trunk/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
URL: http://svn.apache.org/viewvc/hadoop/common/trunk/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?rev=1480809&r1=1480808&r2=1480809&view=diff
==============================================================================
--- hadoop/common/trunk/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 (original)
+++ hadoop/common/trunk/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 Thu May  9 22:15:58 2013
@@ -68,6 +68,7 @@ import org.apache.hadoop.yarn.server.res
 import org.apache.hadoop.yarn.server.resourcemanager.rmapp.attempt.RMAppAttemptState;
 import org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainer;
 import org.apache.hadoop.yarn.server.resourcemanager.rmnode.RMNode;
+import org.apache.hadoop.yarn.server.resourcemanager.scheduler.QueueMetrics;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.ResourceScheduler;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.AppAddedSchedulerEvent;
 import org.apache.hadoop.yarn.server.resourcemanager.scheduler.event.AppRemovedSchedulerEvent;
@@ -128,6 +129,7 @@ public class TestFairScheduler {
   public void tearDown() {
     scheduler = null;
     resourceManager = null;
+    QueueMetrics.clearQueueMetrics();
   }
 
   private Configuration createConfiguration() {
@@ -337,6 +339,13 @@ public class TestFairScheduler {
 
     assertEquals(1024, scheduler.getQueueManager().getQueue("queue1").
       getResourceUsage().getMemory());
+
+    // verify metrics
+    QueueMetrics queue1Metrics = scheduler.getQueueManager().getQueue("queue1")
+        .getMetrics();
+    assertEquals(1024, queue1Metrics.getAllocatedMB());
+    assertEquals(1024, scheduler.getRootQueueMetrics().getAllocatedMB());
+    assertEquals(512, scheduler.getRootQueueMetrics().getAvailableMB());
   }
 
   @Test (timeout = 5000)
@@ -1276,6 +1285,7 @@ public class TestFairScheduler {
     scheduler.handle(updateEvent);
 
     assertEquals(1, app.getLiveContainers().size());
+    assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
     
     // Create request at higher priority
     createSchedulingRequestExistingApplication(1024, 1, attId);
@@ -1291,6 +1301,7 @@ public class TestFairScheduler {
     // Complete container
     scheduler.allocate(attId, new ArrayList<ResourceRequest>(),
         Arrays.asList(containerId));
+    assertEquals(1024, scheduler.getRootQueueMetrics().getAvailableMB());
     
     // Schedule at opening
     scheduler.update();
@@ -1302,6 +1313,7 @@ public class TestFairScheduler {
     for (RMContainer liveContainer : liveContainers) {
       Assert.assertEquals(2, liveContainer.getContainer().getPriority().getPriority());
     }
+    assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
   }
   
   @Test
@@ -1606,4 +1618,24 @@ public class TestFairScheduler {
     assertEquals(1, app.getLiveContainers().size());
     assertEquals(0, app.getReservedContainers().size());
   }
+  
+  @Test
+  public void testRemoveNodeUpdatesRootQueueMetrics() {
+    assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
+    
+    RMNode node1 = MockNodes.newNodeInfo(1, Resources.createResource(1024));
+    NodeAddedSchedulerEvent addEvent = new NodeAddedSchedulerEvent(node1);
+    scheduler.handle(addEvent);
+    
+    assertEquals(1024, scheduler.getRootQueueMetrics().getAvailableMB());
+    scheduler.update(); // update shouldn't change things
+    assertEquals(1024, scheduler.getRootQueueMetrics().getAvailableMB());
+    
+    NodeRemovedSchedulerEvent removeEvent = new NodeRemovedSchedulerEvent(node1);
+    scheduler.handle(removeEvent);
+    
+    assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
+    scheduler.update(); // update shouldn't change things
+    assertEquals(0, scheduler.getRootQueueMetrics().getAvailableMB());
+  }
 }