HDFS-3126. Journal stream from Namenode to BackupNode needs to have timeout. Contributed by Hari Mankude.

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/trunk@1308636 13f79535-47bb-0310-9956-ffa450edef68

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

@@ -109,6 +109,9 @@ Trunk (unreleased changes)
 
     HDFS-3116. Typo in fetchdt error message. (AOE Takashi via atm)
 
+    HDFS-3126. Journal stream from Namenode to BackupNode needs to have 
+    timeout. (Hari Mankude via suresh)
+    
 Release 2.0.0 - UNRELEASED 
 
   INCOMPATIBLE CHANGES

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

@@ -190,7 +190,7 @@ public class NameNodeProxies {
       InetSocketAddress address, Configuration conf, UserGroupInformation ugi)
       throws IOException {
     JournalProtocolPB proxy = (JournalProtocolPB) createNameNodeProxy(address,
-        conf, ugi, JournalProtocolPB.class);
+        conf, ugi, JournalProtocolPB.class, 30000);
     return new JournalProtocolTranslatorPB(proxy);
   }
 
@@ -198,7 +198,7 @@ public class NameNodeProxies {
       createNNProxyWithRefreshAuthorizationPolicyProtocol(InetSocketAddress address,
           Configuration conf, UserGroupInformation ugi) throws IOException {
     RefreshAuthorizationPolicyProtocolPB proxy = (RefreshAuthorizationPolicyProtocolPB)
-        createNameNodeProxy(address, conf, ugi, RefreshAuthorizationPolicyProtocolPB.class);
+        createNameNodeProxy(address, conf, ugi, RefreshAuthorizationPolicyProtocolPB.class, 0);
     return new RefreshAuthorizationPolicyProtocolClientSideTranslatorPB(proxy);
   }
   
@@ -206,7 +206,7 @@ public class NameNodeProxies {
       createNNProxyWithRefreshUserMappingsProtocol(InetSocketAddress address,
           Configuration conf, UserGroupInformation ugi) throws IOException {
     RefreshUserMappingsProtocolPB proxy = (RefreshUserMappingsProtocolPB)
-        createNameNodeProxy(address, conf, ugi, RefreshUserMappingsProtocolPB.class);
+        createNameNodeProxy(address, conf, ugi, RefreshUserMappingsProtocolPB.class, 0);
     return new RefreshUserMappingsProtocolClientSideTranslatorPB(proxy);
   }
 
@@ -214,7 +214,7 @@ public class NameNodeProxies {
       InetSocketAddress address, Configuration conf, UserGroupInformation ugi)
       throws IOException {
     GetUserMappingsProtocolPB proxy = (GetUserMappingsProtocolPB)
-        createNameNodeProxy(address, conf, ugi, GetUserMappingsProtocolPB.class);
+        createNameNodeProxy(address, conf, ugi, GetUserMappingsProtocolPB.class, 0);
     return new GetUserMappingsProtocolClientSideTranslatorPB(proxy);
   }
   
@@ -222,7 +222,7 @@ public class NameNodeProxies {
       InetSocketAddress address, Configuration conf, UserGroupInformation ugi,
       boolean withRetries) throws IOException {
     NamenodeProtocolPB proxy = (NamenodeProtocolPB) createNameNodeProxy(
-        address, conf, ugi, NamenodeProtocolPB.class);
+        address, conf, ugi, NamenodeProtocolPB.class, 0);
     if (withRetries) { // create the proxy with retries
       RetryPolicy timeoutPolicy = RetryPolicies.exponentialBackoffRetry(5, 200,
           TimeUnit.MILLISECONDS);
@@ -244,7 +244,7 @@ public class NameNodeProxies {
       InetSocketAddress address, Configuration conf, UserGroupInformation ugi,
       boolean withRetries) throws IOException {
     ClientNamenodeProtocolPB proxy = (ClientNamenodeProtocolPB) NameNodeProxies
-        .createNameNodeProxy(address, conf, ugi, ClientNamenodeProtocolPB.class);
+        .createNameNodeProxy(address, conf, ugi, ClientNamenodeProtocolPB.class, 0);
     if (withRetries) { // create the proxy with retries
       RetryPolicy createPolicy = RetryPolicies
           .retryUpToMaximumCountWithFixedSleep(5,
@@ -275,11 +275,11 @@ public class NameNodeProxies {
   
   @SuppressWarnings("unchecked")
   private static Object createNameNodeProxy(InetSocketAddress address,
-      Configuration conf, UserGroupInformation ugi, Class xface)
+      Configuration conf, UserGroupInformation ugi, Class xface, int rpcTimeout)
       throws IOException {
     RPC.setProtocolEngine(conf, xface, ProtobufRpcEngine.class);
     Object proxy = RPC.getProxy(xface, RPC.getProtocolVersion(xface), address,
-        ugi, conf, NetUtils.getDefaultSocketFactory(conf));
+        ugi, conf, NetUtils.getDefaultSocketFactory(conf), rpcTimeout);
     return proxy;
   }
 

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

@@ -46,6 +46,7 @@ import org.apache.hadoop.ipc.RPC;
 import org.apache.hadoop.net.NetUtils;
 import org.apache.hadoop.security.UserGroupInformation;
 
+import com.google.common.annotations.VisibleForTesting;
 import com.google.protobuf.BlockingService;
 
 /**
@@ -171,6 +172,12 @@ public class BackupNode extends NameNode {
 
   @Override // NameNode
   public void stop() {
+    stop(true);
+  }
+  
+  @VisibleForTesting
+  void stop(boolean reportError) {
+   
     if(checkpointManager != null) {
       // Prevent from starting a new checkpoint.
       // Checkpoints that has already been started may proceed until 
@@ -180,7 +187,10 @@ public class BackupNode extends NameNode {
       // ClosedByInterruptException.
       checkpointManager.shouldRun = false;
     }
-    if(namenode != null && getRegistration() != null) {
+    
+    // reportError is a test hook to simulate backupnode crashing and not
+    // doing a clean exit w.r.t active namenode
+    if (reportError && namenode != null && getRegistration() != null) {
       // Exclude this node from the list of backup streams on the name-node
       try {
         namenode.errorReport(getRegistration(), NamenodeProtocol.FATAL,

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

@@ -1020,7 +1020,7 @@ public class FSEditLog  {
     
     LOG.info("Registering new backup node: " + bnReg);
     BackupJournalManager bjm = new BackupJournalManager(bnReg, nnReg);
-    journalSet.add(bjm, true);
+    journalSet.add(bjm, false);
   }
   
   synchronized void releaseBackupStream(NamenodeRegistration registration)

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

@@ -185,6 +185,16 @@ public class TestBackupNode {
 
       testBNInSync(cluster, backup, 4);
       assertNotNull(backup.getNamesystem().getFileInfo("/edit-while-bn-down", false));
+      
+      // Trigger an unclean shutdown of the backup node. Backup node will not
+      // unregister from the active when this is done simulating a node crash.
+      backup.stop(false);
+           
+      // do some edits on the active. This should go through without failing.
+      // This will verify that active is still up and can add entries to
+      // master editlog.
+      assertTrue(fileSys.mkdirs(new Path("/edit-while-bn-down-2")));
+      
     } finally {
       LOG.info("Shutting down...");
       if (backup != null) backup.stop();