Revert bad HDFS-6268 commit from branch-2

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1599807 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -20,10 +20,8 @@ package org.apache.hadoop.net;
 import java.util.ArrayList;
 import java.util.List;
 import java.util.Collection;
-import java.util.Collections;
 import java.util.List;
 import java.util.Random;
-import java.util.TreeMap;
 import java.util.concurrent.locks.ReadWriteLock;
 import java.util.concurrent.locks.ReentrantReadWriteLock;
 
@@ -35,9 +33,6 @@ import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.CommonConfigurationKeysPublic;
 import org.apache.hadoop.util.ReflectionUtils;
 
-import com.google.common.base.Preconditions;
-import com.google.common.collect.Lists;
-
 /** The class represents a cluster of computer with a tree hierarchical
  * network topology.
  * For example, a cluster may be consists of many data centers filled 
@@ -673,23 +668,7 @@ public class NetworkTopology {
     return node1.getParent()==node2.getParent();
   }
 
-  private static final ThreadLocal<Random> r = new ThreadLocal<Random>();
+  final protected static Random r = new Random();
-
-  /**
-   * Getter for thread-local Random, which provides better performance than
-   * a shared Random (even though Random is thread-safe).
-   *
-   * @return Thread-local Random.
-   */
-  protected Random getRandom() {
-    Random rand = r.get();
-    if (rand == null) {
-      rand = new Random();
-      r.set(rand);
-    }
-    return rand;
-  }
-
   /** 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>
@@ -739,7 +718,7 @@ public class NetworkTopology {
           "Failed to find datanode (scope=\"" + String.valueOf(scope) +
           "\" excludedScope=\"" + String.valueOf(excludedScope) + "\").");
     }
-    int leaveIndex = getRandom().nextInt(numOfDatanodes);
+    int leaveIndex = r.nextInt(numOfDatanodes);
     return innerNode.getLeaf(leaveIndex, node);
   }
 
@@ -846,79 +825,61 @@ public class NetworkTopology {
     return networkLocation.substring(index);
   }
 
-  /**
+  /** swap two array items */
-   * Returns an integer weight which specifies how far away {node} is away from
+  static protected void swap(Node[] nodes, int i, int j) {
-   * {reader}. A lower value signifies that a node is closer.
+    Node tempNode;
-   * 
+    tempNode = nodes[j];
-   * @param reader Node where data will be read
+    nodes[j] = nodes[i];
-   * @param node Replica of data
+    nodes[i] = tempNode;
-   * @return weight
-   */
-  protected int getWeight(Node reader, Node node) {
-    // 0 is local, 1 is same rack, 2 is off rack
-    // Start off by initializing to off rack
-    int weight = 2;
-    if (reader != null) {
-      if (reader == node) {
-        weight = 0;
-      } else if (isOnSameRack(reader, node)) {
-        weight = 1;
-      }
-    }
-    return weight;
   }
-
+  
-  /**
+  /** Sort nodes array by their distances to <i>reader</i>
-   * Sort nodes array by network distance to <i>reader</i>.
+   * It linearly scans the array, if a local node is found, swap it with
-   * <p/>
+   * the first element of the array.
-   * In a three-level topology, a node can be either local, on the same rack, or
+   * If a local rack node is found, swap it with the first element following
-   * on a different rack from the reader. Sorting the nodes based on network
+   * the local node.
-   * distance from the reader reduces network traffic and improves performance.
+   * If neither local node or local rack node is found, put a random replica
-   * <p/>
+   * location at position 0.
-   * As an additional twist, we also randomize the nodes at each network
+   * It leaves the rest nodes untouched.
-   * distance using the provided random seed. This helps with load balancing
+   * @param reader the node that wishes to read a block from one of the nodes
-   * when there is data skew.
+   * @param nodes the list of nodes containing data for the reader
-   * 
-   * @param reader Node where data will be read
-   * @param nodes Available replicas with the requested data
-   * @param seed Used to seed the pseudo-random generator that randomizes the
-   *          set of nodes at each network distance.
    */
-  public void sortByDistance(Node reader, Node[] nodes, long seed) {
+  public void pseudoSortByDistance( Node reader, Node[] nodes ) {
-    /** Sort weights for the nodes array */
+    int tempIndex = 0;
-    int[] weights = new int[nodes.length];
+    int localRackNode = -1;
-    for (int i=0; i<nodes.length; i++) {
+    if (reader != null ) {
-      weights[i] = getWeight(reader, nodes[i]);
+      //scan the array to find the local node & local rack node
-    }
+      for(int i=0; i<nodes.length; i++) {
-    // Add weight/node pairs to a TreeMap to sort
+        if(tempIndex == 0 && reader == nodes[i]) { //local node
-    TreeMap<Integer, List<Node>> tree = new TreeMap<Integer, List<Node>>();
+          //swap the local node and the node at position 0
-    for (int i=0; i<nodes.length; i++) {
+          if( i != 0 ) {
-      int weight = weights[i];
+            swap(nodes, tempIndex, i);
-      Node node = nodes[i];
+          }
-      List<Node> list = tree.get(weight);
+          tempIndex=1;
-      if (list == null) {
+          if(localRackNode != -1 ) {
-        list = Lists.newArrayListWithExpectedSize(1);
+            if(localRackNode == 0) {
-        tree.put(weight, list);
+              localRackNode = i;
-      }
+            }
-      list.add(node);
+            break;
-    }
+          }
-
+        } else if(localRackNode == -1 && isOnSameRack(reader, nodes[i])) {
-    // Seed is normally the block id
+          //local rack
-    // This means we use the same pseudo-random order for each block, for
+          localRackNode = i;
-    // potentially better page cache usage.
+          if(tempIndex != 0 ) break;
-    Random rand = getRandom();
-    rand.setSeed(seed);
-    int idx = 0;
-    for (List<Node> list: tree.values()) {
-      if (list != null) {
-        Collections.shuffle(list, rand);
-        for (Node n: list) {
-          nodes[idx] = n;
-          idx++;
         }
       }
+
+      // swap the local rack node and the node at position tempIndex
+      if(localRackNode != -1 && localRackNode != tempIndex ) {
+        swap(nodes, tempIndex, localRackNode);
+        tempIndex++;
+      }
+    }
+    
+    // put a random node at position 0 if it is not a local/local-rack node
+    if(tempIndex == 0 && localRackNode == -1 && nodes.length != 0) {
+      swap(nodes, 0, r.nextInt(nodes.length));
     }
-    Preconditions.checkState(idx == nodes.length,
-        "Sorted the wrong number of nodes!");
   }
+  
 }

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

@@ -248,41 +248,25 @@ public class NetworkTopologyWithNodeGroup extends NetworkTopology {
     }
   }
 
-  @Override
+  /** Sort nodes array by their distances to <i>reader</i>
-  protected int getWeight(Node reader, Node node) {
+   * It linearly scans the array, if a local node is found, swap it with
-    // 0 is local, 1 is same node group, 2 is same rack, 3 is off rack
+   * the first element of the array.
-    // Start off by initializing to off rack
+   * If a local node group node is found, swap it with the first element 
-    int weight = 3;
+   * following the local node.
-    if (reader != null) {
+   * If a local rack node is found, swap it with the first element following
-      if (reader == node) {
+   * the local node group node.
-        weight = 0;
+   * If neither local node, node group node or local rack node is found, put a 
-      } else if (isOnSameNodeGroup(reader, node)) {
+   * random replica location at position 0.
-        weight = 1;
+   * It leaves the rest nodes untouched.
-      } else if (isOnSameRack(reader, node)) {
+   * @param reader the node that wishes to read a block from one of the nodes
-        weight = 2;
+   * @param nodes the list of nodes containing data for the reader
-      }
-    }
-    return weight;
-  }
-
-  /**
-   * Sort nodes array by their distances to <i>reader</i>.
-   * <p/>
-   * This is the same as
-   * {@link NetworkTopology#sortByDistance(Node, Node[], long)} except with a
-   * four-level network topology which contains the additional network distance
-   * of a "node group" which is between local and same rack.
-   * 
-   * @param reader Node where data will be read
-   * @param nodes Available replicas with the requested data
-   * @param seed Used to seed the pseudo-random generator that randomizes the
-   *          set of nodes at each network distance.
    */
   @Override
-  public void sortByDistance( Node reader, Node[] nodes, long seed) {
+  public void pseudoSortByDistance( Node reader, Node[] nodes ) {
-    // If reader is not a datanode (not in NetworkTopology tree), we need to
+
-    // replace this reader with a sibling leaf node in tree.
     if (reader != null && !this.contains(reader)) {
+      // if reader is not a datanode (not in NetworkTopology tree), we will 
+      // replace this reader with a sibling leaf node in tree.
       Node nodeGroup = getNode(reader.getNetworkLocation());
       if (nodeGroup != null && nodeGroup instanceof InnerNode) {
         InnerNode parentNode = (InnerNode) nodeGroup;
@@ -292,7 +276,62 @@ public class NetworkTopologyWithNodeGroup extends NetworkTopology {
         return;
       }
     }
-    super.sortByDistance(reader, nodes, seed);
+    int tempIndex = 0;
+    int localRackNode = -1;
+    int localNodeGroupNode = -1;
+    if (reader != null) {  
+      //scan the array to find the local node & local rack node
+      for (int i = 0; i < nodes.length; i++) {
+        if (tempIndex == 0 && reader == nodes[i]) { //local node
+          //swap the local node and the node at position 0
+          if (i != 0) {
+            swap(nodes, tempIndex, i);
+          }
+          tempIndex=1;
+
+          if (localRackNode != -1 && (localNodeGroupNode !=-1)) {
+            if (localRackNode == 0) {
+              localRackNode = i;
+            }
+            if (localNodeGroupNode == 0) {
+              localNodeGroupNode = i;
+            }
+            break;
+          }
+        } else if (localNodeGroupNode == -1 && isOnSameNodeGroup(reader, 
+            nodes[i])) {
+          //local node group
+          localNodeGroupNode = i;
+          // node local and rack local are already found
+          if(tempIndex != 0 && localRackNode != -1) break;
+        } else if (localRackNode == -1 && isOnSameRack(reader, nodes[i])) {
+          localRackNode = i;
+          if (tempIndex != 0 && localNodeGroupNode != -1) break;
+        }
+      }
+
+      // swap the local nodegroup node and the node at position tempIndex
+      if(localNodeGroupNode != -1 && localNodeGroupNode != tempIndex) {
+        swap(nodes, tempIndex, localNodeGroupNode);
+        if (localRackNode == tempIndex) {
+          localRackNode = localNodeGroupNode;
+        }
+        tempIndex++;
+      }
+
+      // swap the local rack node and the node at position tempIndex
+      if(localRackNode != -1 && localRackNode != tempIndex) {
+        swap(nodes, tempIndex, localRackNode);
+        tempIndex++;
+      }
+    }
+
+    // put a random node at position 0 if there is not a local/local-nodegroup/
+    // local-rack node
+    if (tempIndex == 0 && localNodeGroupNode == -1 && localRackNode == -1
+        && nodes.length != 0) {
+      swap(nodes, 0, r.nextInt(nodes.length));
+    }
   }
 
   /** InnerNodeWithNodeGroup represents a switch/router of a data center, rack

hadoop-common-project/hadoop-common/src/test/java/org/apache/hadoop/net/TestNetworkTopologyWithNodeGroup.java

@@ -96,7 +96,7 @@ public class TestNetworkTopologyWithNodeGroup {
   }
 
   @Test
-  public void testSortByDistance() throws Exception {
+  public void testPseudoSortByDistance() throws Exception {
     NodeBase[] testNodes = new NodeBase[4];
 
     // array contains both local node, local node group & local rack node
@@ -104,7 +104,7 @@ public class TestNetworkTopologyWithNodeGroup {
     testNodes[1] = dataNodes[2];
     testNodes[2] = dataNodes[3];
     testNodes[3] = dataNodes[0];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[1]);
     assertTrue(testNodes[2] == dataNodes[2]);
@@ -115,7 +115,7 @@ public class TestNetworkTopologyWithNodeGroup {
     testNodes[1] = dataNodes[4];
     testNodes[2] = dataNodes[1];
     testNodes[3] = dataNodes[0];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[1]);
 
@@ -124,7 +124,7 @@ public class TestNetworkTopologyWithNodeGroup {
     testNodes[1] = dataNodes[3];
     testNodes[2] = dataNodes[2];
     testNodes[3] = dataNodes[0];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[2]);
 
@@ -133,7 +133,7 @@ public class TestNetworkTopologyWithNodeGroup {
     testNodes[1] = dataNodes[7];
     testNodes[2] = dataNodes[2];
     testNodes[3] = dataNodes[0];
-    cluster.sortByDistance(computeNode, testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(computeNode, testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[2]);
   }

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -147,9 +147,6 @@ Release 2.5.0 - UNRELEASED
     HDFS-6109 let sync_file_range() system call run in background
     (Liang Xie via stack)
 
-    HDFS-6268. Better sorting in NetworkTopology#pseudoSortByDistance when
-    no local node is found. (wang)
-
   OPTIMIZATIONS
 
     HDFS-6214. Webhdfs has poor throughput for files >2GB (daryn)

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

@@ -351,8 +351,7 @@ public class DatanodeManager {
         DFSUtil.DECOM_COMPARATOR;
         
     for (LocatedBlock b : locatedblocks) {
-      networktopology.sortByDistance(client, b.getLocations(), b
+      networktopology.pseudoSortByDistance(client, b.getLocations());
-          .getBlock().getBlockId());
       // Move decommissioned/stale datanodes to the bottom
       Arrays.sort(b.getLocations(), comparator);
     }

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java

@@ -1618,11 +1618,9 @@ public class FSNamesystem implements Namesystem, FSClusterStats,
       blockManager.getDatanodeManager().sortLocatedBlocks(
           clientMachine, blocks.getLocatedBlocks());
       
-      // lastBlock is not part of getLocatedBlocks(), might need to sort it too
       LocatedBlock lastBlock = blocks.getLastLocatedBlock();
       if (lastBlock != null) {
-        ArrayList<LocatedBlock> lastBlockList =
+        ArrayList<LocatedBlock> lastBlockList = new ArrayList<LocatedBlock>();
-            Lists.newArrayListWithCapacity(1);
         lastBlockList.add(lastBlock);
         blockManager.getDatanodeManager().sortLocatedBlocks(
                               clientMachine, lastBlockList);

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestGetBlocks.java

@@ -169,9 +169,6 @@ public class TestGetBlocks {
       if (stm != null) {
         stm.close();
       }
-      if (client != null) {
-        client.close();
-      }
       cluster.shutdown();
     }
   }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/snapshot/TestSnapshotFileLength.java

@@ -143,10 +143,10 @@ public class TestSnapshotFileLength {
 
     // Make sure we can read the entire file via its non-snapshot path.
     fileStatus = hdfs.getFileStatus(file1);
-    assertEquals("Unexpected file length", BLOCKSIZE * 2, fileStatus.getLen());
+    assertEquals(fileStatus.getLen(), BLOCKSIZE * 2);
     fis = hdfs.open(file1);
     bytesRead = fis.read(buffer, 0, buffer.length);
-    assertEquals("Unexpected # bytes read", BLOCKSIZE * 2, bytesRead);
+    assertEquals(bytesRead, BLOCKSIZE * 2);
     fis.close();
 
     Path file1snap1 =
@@ -156,23 +156,21 @@ public class TestSnapshotFileLength {
     assertEquals(fileStatus.getLen(), BLOCKSIZE);
     // Make sure we can only read up to the snapshot length.
     bytesRead = fis.read(buffer, 0, buffer.length);
-    assertEquals("Unexpected # bytes read", BLOCKSIZE, bytesRead);
+    assertEquals(bytesRead, BLOCKSIZE);
     fis.close();
 
-    PrintStream outBackup = System.out;
+    PrintStream psBackup = System.out;
-    PrintStream errBackup = System.err;
     ByteArrayOutputStream bao = new ByteArrayOutputStream();
     System.setOut(new PrintStream(bao));
     System.setErr(new PrintStream(bao));
     // Make sure we can cat the file upto to snapshot length
     FsShell shell = new FsShell();
-    try {
+    try{
       ToolRunner.run(conf, shell, new String[] { "-cat",
       "/TestSnapshotFileLength/sub1/.snapshot/snapshot1/file1" });
-      assertEquals("Unexpected # bytes from -cat", BLOCKSIZE, bao.size());
+      assertEquals(bao.size(), BLOCKSIZE);
-    } finally {
+    }finally{
-      System.setOut(outBackup);
+      System.setOut(psBackup);
-      System.setErr(errBackup);
     }
   }
 }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/net/TestNetworkTopology.java

@@ -54,8 +54,7 @@ public class TestNetworkTopology {
         DFSTestUtil.getDatanodeDescriptor("4.4.4.4", "/d1/r2"),
         DFSTestUtil.getDatanodeDescriptor("5.5.5.5", "/d1/r2"),
         DFSTestUtil.getDatanodeDescriptor("6.6.6.6", "/d2/r3"),
-        DFSTestUtil.getDatanodeDescriptor("7.7.7.7", "/d2/r3"),
+        DFSTestUtil.getDatanodeDescriptor("7.7.7.7", "/d2/r3")
-        DFSTestUtil.getDatanodeDescriptor("8.8.8.8", "/d2/r3")
     };
     for (int i = 0; i < dataNodes.length; i++) {
       cluster.add(dataNodes[i]);
@@ -118,14 +117,14 @@ public class TestNetworkTopology {
   }
 
   @Test
-  public void testSortByDistance() throws Exception {
+  public void testPseudoSortByDistance() throws Exception {
     DatanodeDescriptor[] testNodes = new DatanodeDescriptor[3];
     
     // array contains both local node & local rack node
     testNodes[0] = dataNodes[1];
     testNodes[1] = dataNodes[2];
     testNodes[2] = dataNodes[0];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[1]);
     assertTrue(testNodes[2] == dataNodes[2]);
@@ -134,7 +133,7 @@ public class TestNetworkTopology {
     testNodes[0] = dataNodes[1];
     testNodes[1] = dataNodes[3];
     testNodes[2] = dataNodes[0];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[0]);
     assertTrue(testNodes[1] == dataNodes[1]);
     assertTrue(testNodes[2] == dataNodes[3]);
@@ -143,50 +142,21 @@ public class TestNetworkTopology {
     testNodes[0] = dataNodes[5];
     testNodes[1] = dataNodes[3];
     testNodes[2] = dataNodes[1];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
     assertTrue(testNodes[0] == dataNodes[1]);
     assertTrue(testNodes[1] == dataNodes[3]);
     assertTrue(testNodes[2] == dataNodes[5]);
-
+    
     // array contains local rack node which happens to be in position 0
     testNodes[0] = dataNodes[1];
     testNodes[1] = dataNodes[5];
     testNodes[2] = dataNodes[3];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEADBEEF);
+    cluster.pseudoSortByDistance(dataNodes[0], testNodes );
-    assertTrue(testNodes[0] == dataNodes[1]);
+    // peudoSortByDistance does not take the "data center" layer into consideration 
-    assertTrue(testNodes[1] == dataNodes[3]);
-    assertTrue(testNodes[2] == dataNodes[5]);
-
-    // Same as previous, but with a different random seed to test randomization
-    testNodes[0] = dataNodes[1];
-    testNodes[1] = dataNodes[5];
-    testNodes[2] = dataNodes[3];
-    cluster.sortByDistance(dataNodes[0], testNodes, 0xDEAD);
-    // sortByDistance does not take the "data center" layer into consideration
     // and it doesn't sort by getDistance, so 1, 5, 3 is also valid here
     assertTrue(testNodes[0] == dataNodes[1]);
     assertTrue(testNodes[1] == dataNodes[5]);
     assertTrue(testNodes[2] == dataNodes[3]);
-
-    // Array is just local rack nodes
-    // Expect a random first node depending on the seed (normally the block ID).
-    DatanodeDescriptor first = null;
-    boolean foundRandom = false;
-    for (int i=5; i<=7; i++) {
-      testNodes[0] = dataNodes[5];
-      testNodes[1] = dataNodes[6];
-      testNodes[2] = dataNodes[7];
-      cluster.sortByDistance(dataNodes[i], testNodes, 0xBEADED+i);
-      if (first == null) {
-        first = testNodes[0];
-      } else {
-        if (first != testNodes[0]) {
-          foundRandom = true;
-          break;
-        }
-      }
-    }
-    assertTrue("Expected to find a different first location", foundRandom);
   }
   
   @Test