svn merge -c 1347867 from trunk for HADOOP-8469. Make NetworkTopology class pluggable.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1488838 13f79535-47bb-0310-9956-ffa450edef68

hadoop-common-project/hadoop-common/CHANGES.txt

@@ -32,6 +32,9 @@ Release 2.1.0-beta - UNRELEASED
     Azure environments. (See breakdown of tasks below for subtasks and
     contributors)
 
+    HADOOP-8469. Make NetworkTopology class pluggable.  (Junping Du via
+    szetszwo)
+
   IMPROVEMENTS
 
     HADOOP-9253. Capture ulimit info in the logs at service start time.

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/CommonConfigurationKeysPublic.java

@@ -63,7 +63,9 @@ public class CommonConfigurationKeysPublic {
   /** See <a href="{@docRoot}/../core-default.html">core-default.xml</a> */
   public static final String  NET_TOPOLOGY_NODE_SWITCH_MAPPING_IMPL_KEY =
     "net.topology.node.switch.mapping.impl";
-  
+  /** See <a href="{@docRoot}/../core-default.html">core-default.xml</a> */
+  public static final String  NET_TOPOLOGY_IMPL_KEY =
+    "net.topology.impl";
   /** See <a href="{@docRoot}/../core-default.html">core-default.xml</a> */
   public static final String  NET_TOPOLOGY_TABLE_MAPPING_FILE_KEY =
     "net.topology.table.file.name";

hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/net/NetworkTopology.java

@@ -20,6 +20,7 @@ package org.apache.hadoop.net;
 import java.util.ArrayList;
 import java.util.List;
 import java.util.Collection;
+import java.util.List;
 import java.util.Random;
 import java.util.concurrent.locks.ReadWriteLock;
 import java.util.concurrent.locks.ReentrantReadWriteLock;
@@ -56,8 +57,8 @@ public class NetworkTopology {
   /** InnerNode represents a switch/router of a data center or rack.
    * Different from a leaf node, it has non-null children.
    */
-  private class InnerNode extends NodeBase {
-    private ArrayList<Node> children=new ArrayList<Node>();
+  static class InnerNode extends NodeBase {
+    protected List<Node> children=new ArrayList<Node>();
     private int numOfLeaves;
         
     /** Construct an InnerNode from a path-like string */
@@ -77,7 +78,7 @@ public class NetworkTopology {
     }
         
     /** @return its children */
-    Collection<Node> getChildren() {return children;}
+    List<Node> getChildren() {return children;}
         
     /** @return the number of children this node has */
     int getNumOfChildren() {
@@ -183,7 +184,23 @@ public class NetworkTopology {
         }
       }
     }
-        
+
+    /**
+     * Creates a parent node to be added to the list of children.  
+     * Creates a node using the InnerNode four argument constructor specifying 
+     * the name, location, parent, and level of this node.
+     * 
+     * <p>To be overridden in subclasses for specific InnerNode implementations,
+     * as alternative to overriding the full {@link #add(Node)} method.
+     * 
+     * @param parentName The name of the parent node
+     * @return A new inner node
+     * @see InnerNode#InnerNode(String, String, InnerNode, int)
+     */
+    protected InnerNode createParentNode(String parentName) {
+      return new InnerNode(parentName, getPath(this), this, this.getLevel()+1);
+    }
+
     /** Remove node <i>n</i> from the subtree of this node
      * @param n node to be deleted 
      * @return true if the node is deleted; false otherwise
@@ -264,7 +281,7 @@ public class NetworkTopology {
      * @param excludedNode an excluded node (can be null)
      * @return
      */
-    private Node getLeaf(int leafIndex, Node excludedNode) {
+    Node getLeaf(int leafIndex, Node excludedNode) {
       int count=0;
       // check if the excluded node a leaf
       boolean isLeaf =
@@ -309,7 +326,21 @@ public class NetworkTopology {
         return null;
       }
     }
-        
+    
+    /**
+      * Determine if children a leaves, default implementation calls {@link #isRack()}
+      * <p>To be overridden in subclasses for specific InnerNode implementations,
+      * as alternative to overriding the full {@link #getLeaf(int, Node)} method.
+      * 
+      * @return true if children are leaves, false otherwise
+      */
+    protected boolean areChildrenLeaves() {
+      return isRack();
+    }
+
+    /**
+     * Get number of leaves.
+     */
     int getNumOfLeaves() {
       return numOfLeaves;
     }
@@ -318,18 +349,18 @@ public class NetworkTopology {
   /**
    * the root cluster map
    */
-  InnerNode clusterMap = new InnerNode(InnerNode.ROOT);
+  InnerNode clusterMap;
   /** Depth of all leaf nodes */
   private int depthOfAllLeaves = -1;
   /** rack counter */
-  private int numOfRacks = 0;
+  protected int numOfRacks = 0;
   /** the lock used to manage access */
-  private ReadWriteLock netlock;
-    
+  protected ReadWriteLock netlock = new ReentrantReadWriteLock();
+
   public NetworkTopology() {
-    netlock = new ReentrantReadWriteLock();
+    clusterMap = new InnerNode(InnerNode.ROOT);
   }
-    
+
   /** Add a leaf node
    * Update node counter & rack counter if necessary
    * @param node node to be added; can be null
@@ -353,7 +384,7 @@ public class NetworkTopology {
             "You cannot have a rack and a non-rack node at the same " +
             "level of the network topology.");
       }
-      Node rack = getNode(node.getNetworkLocation());
+      Node rack = getNodeForNetworkLocation(node);
       if (rack != null && !(rack instanceof InnerNode)) {
         throw new IllegalArgumentException("Unexpected data node " 
                                            + node.toString() 
@@ -377,7 +408,26 @@ public class NetworkTopology {
       netlock.writeLock().unlock();
     }
   }
-    
+  
+  /**
+   * Return a reference to the node given its string representation.
+   * Default implementation delegates to {@link #getNode(String)}.
+   * 
+   * <p>To be overridden in subclasses for specific NetworkTopology 
+   * implementations, as alternative to overriding the full {@link #add(Node)}
+   *  method.
+   * 
+   * @param node The string representation of this node's network location is
+   * used to retrieve a Node object. 
+   * @return a reference to the node; null if the node is not in the tree
+   * 
+   * @see #add(Node)
+   * @see #getNode(String)
+   */
+  protected Node getNodeForNetworkLocation(Node node) {
+    return getNode(node.getNetworkLocation());
+  }
+  
   /**
    * Given a string representation of a rack, return its children
    * @param loc a path-like string representation of a rack
@@ -426,7 +476,7 @@ public class NetworkTopology {
       netlock.writeLock().unlock();
     }
   }
-       
+
   /** Check if the tree contains node <i>node</i>
    * 
    * @param node a node
@@ -466,7 +516,21 @@ public class NetworkTopology {
       netlock.readLock().unlock();
     }
   }
-    
+  
+  /** Given a string representation of a rack for a specific network
+   *  location
+   * 
+   * To be overridden in subclasses for specific NetworkTopology 
+   * implementations, as alternative to overriding the full 
+   * {@link #getRack(String)} method.
+   * @param loc
+   *          a path-like string representation of a network location
+   * @return a rack string
+   */
+  public String getRack(String loc) {
+    return loc;
+  }
+  
   /** @return the total number of racks */
   public int getNumOfRacks() {
     netlock.readLock().lock();
@@ -476,7 +540,7 @@ public class NetworkTopology {
       netlock.readLock().unlock();
     }
   }
-    
+
   /** @return the total number of leaf nodes */
   public int getNumOfLeaves() {
     netlock.readLock().lock();
@@ -486,7 +550,7 @@ public class NetworkTopology {
       netlock.readLock().unlock();
     }
   }
-    
+
   /** Return the distance between two nodes
    * It is assumed that the distance from one node to its parent is 1
    * The distance between two nodes is calculated by summing up their distances
@@ -532,8 +596,8 @@ public class NetworkTopology {
       return Integer.MAX_VALUE;
     }
     return dis+2;
-  } 
-    
+  }
+
   /** Check if two nodes are on the same rack
    * @param node1 one node (can be null)
    * @param node2 another node (can be null)
@@ -548,13 +612,44 @@ public class NetworkTopology {
       
     netlock.readLock().lock();
     try {
-      return node1.getParent()==node2.getParent();
+      return isSameParents(node1, node2);
     } finally {
       netlock.readLock().unlock();
     }
   }
-    
-  final private static Random r = new Random();
+  
+  /**
+   * Check if network topology is aware of NodeGroup
+   */
+  public boolean isNodeGroupAware() {
+    return false;
+  }
+  
+  /** 
+   * Return false directly as not aware of NodeGroup, to be override in sub-class
+   */
+  public boolean isOnSameNodeGroup(Node node1, Node node2) {
+    return false;
+  }
+
+  /**
+   * Compare the parents of each node for equality
+   * 
+   * <p>To be overridden in subclasses for specific NetworkTopology 
+   * implementations, as alternative to overriding the full 
+   * {@link #isOnSameRack(Node, Node)} method.
+   * 
+   * @param node1 the first node to compare
+   * @param node2 the second node to compare
+   * @return true if their parents are equal, false otherwise
+   * 
+   * @see #isOnSameRack(Node, Node)
+   */
+  protected boolean isSameParents(Node node1, Node node2) {
+    return node1.getParent()==node2.getParent();
+  }
+
+  final protected static Random r = new Random();
   /** randomly choose one node from <i>scope</i>
    * if scope starts with ~, choose one from the all nodes except for the
    * ones in <i>scope</i>; otherwise, choose one from <i>scope</i>
@@ -573,7 +668,7 @@ public class NetworkTopology {
       netlock.readLock().unlock();
     }
   }
-    
+
   private Node chooseRandom(String scope, String excludedScope){
     if (excludedScope != null) {
       if (scope.startsWith(excludedScope)) {
@@ -602,7 +697,25 @@ public class NetworkTopology {
     int leaveIndex = r.nextInt(numOfDatanodes);
     return innerNode.getLeaf(leaveIndex, node);
   }
-       
+
+  /** return leaves in <i>scope</i>
+   * @param scope a path string
+   * @return leaves nodes under specific scope
+   */
+  public List<Node> getLeaves(String scope) {
+    Node node = getNode(scope);
+    List<Node> leafNodes = new ArrayList<Node>();
+    if (!(node instanceof InnerNode)) {
+      leafNodes.add(node);
+    } else {
+      InnerNode innerNode = (InnerNode) node;
+      for (int i=0;i<innerNode.getNumOfLeaves();i++) {
+        leafNodes.add(innerNode.getLeaf(i, null));
+      }
+    }
+    return leafNodes;
+  }
+
   /** return the number of leaves in <i>scope</i> but not in <i>excludedNodes</i>
    * if scope starts with ~, return the number of nodes that are not
    * in <i>scope</i> and <i>excludedNodes</i>; 
@@ -642,7 +755,7 @@ public class NetworkTopology {
       netlock.readLock().unlock();
     }
   }
-    
+
   /** convert a network tree to a string */
   @Override
   public String toString() {
@@ -664,13 +777,12 @@ public class NetworkTopology {
     return tree.toString();
   }
 
-  /* swap two array items */
-  static private void swap(Node[] nodes, int i, int j) {
+  /** swap two array items */
+  static protected void swap(Node[] nodes, int i, int j) {
     Node tempNode;
     tempNode = nodes[j];
     nodes[j] = nodes[i];
     nodes[i] = tempNode;
-    
   }
   
   /** Sort nodes array by their distances to <i>reader</i>
@@ -721,4 +833,5 @@ public class NetworkTopology {
       swap(nodes, 0, r.nextInt(nodes.length));
     }
   }
+  
 }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/protocol/DatanodeInfo.java

@@ -204,10 +204,10 @@ public class DatanodeInfo extends DatanodeID implements Node {
     this.xceiverCount = xceiverCount; 
   }
 
-  /** rack name */
+  /** network location */
   public synchronized String getNetworkLocation() {return location;}
     
-  /** Sets the rack name */
+  /** Sets the network location */
   public synchronized void setNetworkLocation(String location) {
     this.location = NodeBase.normalize(location);
   }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/DatanodeManager.java

@@ -39,6 +39,7 @@ import org.apache.hadoop.HadoopIllegalArgumentException;
 import org.apache.hadoop.classification.InterfaceAudience;
 import org.apache.hadoop.classification.InterfaceStability;
 import org.apache.hadoop.conf.Configuration;
+import org.apache.hadoop.fs.CommonConfigurationKeysPublic;
 import org.apache.hadoop.hdfs.DFSConfigKeys;
 import org.apache.hadoop.hdfs.DFSUtil;
 import org.apache.hadoop.hdfs.HdfsConfiguration;
@@ -113,7 +114,7 @@ public class DatanodeManager {
       = new TreeMap<String, DatanodeDescriptor>();
 
   /** Cluster network topology */
-  private final NetworkTopology networktopology = new NetworkTopology();
+  private final NetworkTopology networktopology;
 
   /** Host names to datanode descriptors mapping. */
   private final Host2NodesMap host2DatanodeMap = new Host2NodesMap();
@@ -161,6 +162,12 @@ public class DatanodeManager {
       final Configuration conf) throws IOException {
     this.namesystem = namesystem;
     this.blockManager = blockManager;
+    
+    Class<? extends NetworkTopology> networkTopologyClass =
+        conf.getClass(CommonConfigurationKeysPublic.NET_TOPOLOGY_IMPL_KEY,
+            NetworkTopology.class, NetworkTopology.class);
+    networktopology = (NetworkTopology) ReflectionUtils.newInstance(
+        networkTopologyClass, conf);
 
     this.heartbeatManager = new HeartbeatManager(namesystem, blockManager, conf);
 
@@ -291,7 +298,16 @@ public class DatanodeManager {
   public void sortLocatedBlocks(final String targethost,
       final List<LocatedBlock> locatedblocks) {
     //sort the blocks
-    final DatanodeDescriptor client = getDatanodeByHost(targethost);
+    // As it is possible for the separation of node manager and datanode, 
+    // here we should get node but not datanode only .
+    Node client = getDatanodeByHost(targethost);
+    if (client == null) {
+      List<String> hosts = new ArrayList<String> (1);
+      hosts.add(targethost);
+      String rName = dnsToSwitchMapping.resolve(hosts).get(0);
+      if (rName != null)
+        client = new NodeBase(rName + NodeBase.PATH_SEPARATOR_STR + targethost);
+    }
     
     Comparator<DatanodeInfo> comparator = avoidStaleDataNodesForRead ?
         new DFSUtil.DecomStaleComparator(staleInterval) : 
@@ -301,7 +317,7 @@ public class DatanodeManager {
       networktopology.pseudoSortByDistance(client, b.getLocations());
       // Move decommissioned/stale datanodes to the bottom
       Arrays.sort(b.getLocations(), comparator);
-    }    
+    }
   }
   
   CyclicIteration<String, DatanodeDescriptor> getDatanodeCyclicIteration(