HDFS-6378. NFS registration should timeout instead of hanging when portmap/rpcbind is not available. Contributed by Abhiraj Butala

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1610543 13f79535-47bb-0310-9956-ffa450edef68

hadoop-common-project/hadoop-nfs/src/main/java/org/apache/hadoop/mount/MountdBase.java

@@ -19,12 +19,16 @@ package org.apache.hadoop.mount;
 
 import java.io.IOException;
 
+import org.apache.commons.logging.Log;
+import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.oncrpc.RpcProgram;
 import org.apache.hadoop.oncrpc.SimpleTcpServer;
 import org.apache.hadoop.oncrpc.SimpleUdpServer;
 import org.apache.hadoop.portmap.PortmapMapping;
 import org.apache.hadoop.util.ShutdownHookManager;
 
+import static org.apache.hadoop.util.ExitUtil.terminate;
+
 /**
  * Main class for starting mountd daemon. This daemon implements the NFS
  * mount protocol. When receiving a MOUNT request from an NFS client, it checks
@@ -33,6 +37,7 @@ import org.apache.hadoop.util.ShutdownHookManager;
  * handle for requested directory and returns it to the client.
  */
 abstract public class MountdBase {
+  public static final Log LOG = LogFactory.getLog(MountdBase.class);
   private final RpcProgram rpcProgram;
   private int udpBoundPort; // Will set after server starts
   private int tcpBoundPort; // Will set after server starts
@@ -40,11 +45,11 @@ abstract public class MountdBase {
   public RpcProgram getRpcProgram() {
     return rpcProgram;
   }
-  
+
   /**
    * Constructor
    * @param program
-   * @throws IOException 
+   * @throws IOException
    */
   public MountdBase(RpcProgram program) throws IOException {
     rpcProgram = program;
@@ -74,11 +79,16 @@ abstract public class MountdBase {
     if (register) {
       ShutdownHookManager.get().addShutdownHook(new Unregister(),
           SHUTDOWN_HOOK_PRIORITY);
-      rpcProgram.register(PortmapMapping.TRANSPORT_UDP, udpBoundPort);
-      rpcProgram.register(PortmapMapping.TRANSPORT_TCP, tcpBoundPort);
+      try {
+        rpcProgram.register(PortmapMapping.TRANSPORT_UDP, udpBoundPort);
+        rpcProgram.register(PortmapMapping.TRANSPORT_TCP, tcpBoundPort);
+      } catch (Throwable e) {
+        LOG.fatal("Failed to start the server. Cause:", e);
+        terminate(1, e);
+      }
     }
   }
-  
+
   /**
    * Priority of the mountd shutdown hook.
    */
@@ -91,5 +101,5 @@ abstract public class MountdBase {
       rpcProgram.unregister(PortmapMapping.TRANSPORT_TCP, tcpBoundPort);
     }
   }
-  
+
 }

hadoop-common-project/hadoop-nfs/src/main/java/org/apache/hadoop/nfs/nfs3/Nfs3Base.java

@@ -25,6 +25,8 @@ import org.apache.hadoop.oncrpc.SimpleTcpServer;
 import org.apache.hadoop.portmap.PortmapMapping;
 import org.apache.hadoop.util.ShutdownHookManager;
 
+import static org.apache.hadoop.util.ExitUtil.terminate;
+
 /**
  * Nfs server. Supports NFS v3 using {@link RpcProgram}.
  * Currently Mountd program is also started inside this class.
@@ -34,7 +36,7 @@ public abstract class Nfs3Base {
   public static final Log LOG = LogFactory.getLog(Nfs3Base.class);
   private final RpcProgram rpcProgram;
   private int nfsBoundPort; // Will set after server starts
-    
+
   public RpcProgram getRpcProgram() {
     return rpcProgram;
   }
@@ -46,11 +48,16 @@ public abstract class Nfs3Base {
 
   public void start(boolean register) {
     startTCPServer(); // Start TCP server
-    
+
     if (register) {
       ShutdownHookManager.get().addShutdownHook(new Unregister(),
           SHUTDOWN_HOOK_PRIORITY);
-      rpcProgram.register(PortmapMapping.TRANSPORT_TCP, nfsBoundPort);
+      try {
+        rpcProgram.register(PortmapMapping.TRANSPORT_TCP, nfsBoundPort);
+      } catch (Throwable e) {
+        LOG.fatal("Failed to start the server. Cause:", e);
+        terminate(1, e);
+      }
     }
   }
 
@@ -61,7 +68,7 @@ public abstract class Nfs3Base {
     tcpServer.run();
     nfsBoundPort = tcpServer.getBoundPort();
   }
-  
+
   /**
    * Priority of the nfsd shutdown hook.
    */

hadoop-common-project/hadoop-nfs/src/main/java/org/apache/hadoop/oncrpc/RpcProgram.java

@@ -131,7 +131,7 @@ public abstract class RpcProgram extends SimpleChannelUpstreamHandler {
     } catch (IOException e) {
       String request = set ? "Registration" : "Unregistration";
       LOG.error(request + " failure with " + host + ":" + port
-          + ", portmap entry: " + mapEntry, e);
+          + ", portmap entry: " + mapEntry);
       throw new RuntimeException(request + " failure", e);
     }
   }

hadoop-common-project/hadoop-nfs/src/main/java/org/apache/hadoop/oncrpc/SimpleUdpClient.java

@@ -60,6 +60,7 @@ public class SimpleUdpClient {
       DatagramPacket sendPacket = new DatagramPacket(sendData, sendData.length,
           IPAddress, port);
       socket.send(sendPacket);
+      socket.setSoTimeout(500);
       DatagramPacket receivePacket = new DatagramPacket(receiveData,
           receiveData.length);
       socket.receive(receivePacket);

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

@@ -825,6 +825,9 @@ Release 2.5.0 - UNRELEASED
     HDFS-6647. Edit log corruption when pipeline recovery occurs for deleted 
     file present in snapshot (kihwal)
 
+    HDFS-6378. NFS registration should timeout instead of hanging when
+    portmap/rpcbind is not available (Abhiraj Butala via brandonli)
+
   BREAKDOWN OF HDFS-2006 SUBTASKS AND RELATED JIRAS
 
     HDFS-6299. Protobuf for XAttr and client-side implementation. (Yi Liu via umamahesh)