HDFS-4946. Allow preferLocalNode in BlockPlacementPolicyDefault to be configurable. Contributed by James Kinley and Nathan Roberts.

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

@@ -2811,6 +2811,9 @@ Release 2.7.3 - UNRELEASED
     HDFS-9669. TcpPeerServer should respect ipc.server.listen.queue.size
     (Elliot Clark via cmccabe)
 
+    HDFS-4946. Allow preferLocalNode in BlockPlacementPolicyDefault to be
+    configurable (James Kinley and Nathan Roberts via kihwal)
+
   OPTIMIZATIONS
 
   BUG FIXES

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/DFSConfigKeys.java

@@ -663,6 +663,9 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
       "dfs.namenode.available-space-block-placement-policy.balanced-space-preference-fraction";
   public static final float   DFS_NAMENODE_AVAILABLE_SPACE_BLOCK_PLACEMENT_POLICY_BALANCED_SPACE_PREFERENCE_FRACTION_DEFAULT =
       0.6f;
+  public static final String  DFS_NAMENODE_BLOCKPLACEMENTPOLICY_DEFAULT_PREFER_LOCAL_NODE_KEY =
+      "dfs.namenode.block-placement-policy.default.prefer-local-node";
+  public static final boolean  DFS_NAMENODE_BLOCKPLACEMENTPOLICY_DEFAULT_PREFER_LOCAL_NODE_DEFAULT = true;
 
   public static final String DFS_BLOCK_LOCAL_PATH_ACCESS_USER_KEY = "dfs.block.local-path-access.user";
   public static final String DFS_DOMAIN_SOCKET_PATH_KEY =

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

@@ -59,7 +59,7 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
 
   protected boolean considerLoad; 
   protected double considerLoadFactor;
-  private boolean preferLocalNode = true;
+  private boolean preferLocalNode;
   protected NetworkTopology clusterMap;
   protected Host2NodesMap host2datanodeMap;
   private FSClusterStats stats;
@@ -95,6 +95,11 @@ public class BlockPlacementPolicyDefault extends BlockPlacementPolicy {
     this.staleInterval = conf.getLong(
         DFSConfigKeys.DFS_NAMENODE_STALE_DATANODE_INTERVAL_KEY, 
         DFSConfigKeys.DFS_NAMENODE_STALE_DATANODE_INTERVAL_DEFAULT);
+    this.preferLocalNode = conf.getBoolean(
+        DFSConfigKeys.
+            DFS_NAMENODE_BLOCKPLACEMENTPOLICY_DEFAULT_PREFER_LOCAL_NODE_KEY,
+        DFSConfigKeys.
+            DFS_NAMENODE_BLOCKPLACEMENTPOLICY_DEFAULT_PREFER_LOCAL_NODE_DEFAULT);
   }
 
   @Override

hadoop-hdfs-project/hadoop-hdfs/src/main/resources/hdfs-default.xml

@@ -921,6 +921,18 @@
   </description>
 </property>
 
+<property>
+  <name>dfs.namenode.block-placement-policy.default.prefer-local-node</name>
+  <value>true</value>
+  <description>Controls how the default block placement policy places
+  the first replica of a block. When true, it will prefer the node where
+  the client is running.  When false, it will prefer a node in the same rack
+  as the client. Setting to false avoids situations where entire copies of
+  large files end up on a single node, thus creating hotspots.
+  </description>
+</property>
+
+
 <property>
   <name>dfs.stream-buffer-size</name>
   <value>4096</value>