HDFS-13183. Addendum: Standby NameNode process getBlocks request to reduce Active load. Contributed by Xiaoqiao He.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/balancer/Balancer.java

@@ -708,12 +708,12 @@ static private int doBalance(Collection<URI> namenodes,
     System.out.println("Time Stamp               Iteration#  Bytes Already Moved  Bytes Left To Move  Bytes Being Moved");
     
     List<NameNodeConnector> connectors = Collections.emptyList();
-    boolean done = false;
-    for(int iteration = 0; !done; iteration++) {
     try {
       connectors = NameNodeConnector.newNameNodeConnectors(namenodes, nsIds,
           Balancer.class.getSimpleName(), BALANCER_ID_PATH, conf,
           p.getMaxIdleIteration());
+      boolean done = false;
+      for(int iteration = 0; !done; iteration++) {
         done = true;
         Collections.shuffle(connectors);
         for(NameNodeConnector nnc : connectors) {
@@ -741,12 +741,12 @@ static private int doBalance(Collection<URI> namenodes,
         if (!done) {
           Thread.sleep(sleeptime);
         }
+      }
     } finally {
       for(NameNodeConnector nnc : connectors) {
         IOUtils.cleanupWithLogger(LOG, nnc);
       }
     }
-    }
     return ExitStatus.SUCCESS.getExitCode();
   }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/balancer/NameNodeConnector.java

@@ -147,12 +147,13 @@ public static void checkOtherInstanceRunning(boolean toCheck) {
 
   private final BalancerProtocols namenode;
   /**
-   * If set balancerShouldRequestStandby true, Balancer will getBlocks from
+   * If set requestToStandby true, Balancer will getBlocks from
    * Standby NameNode only and it can reduce the performance impact of Active
    * NameNode, especially in a busy HA mode cluster.
    */
-  private boolean balancerShouldRequestStandby;
+  private boolean requestToStandby;
-  private NamenodeProtocol standbyNameNode;
+  private String nsId;
+  private Configuration config;
   private final KeyManager keyManager;
   final AtomicBoolean fallbackToSimpleAuth = new AtomicBoolean(false);
 
@@ -188,10 +189,10 @@ public NameNodeConnector(String name, URI nameNodeUri, Path idPath,
 
     this.namenode = NameNodeProxies.createProxy(conf, nameNodeUri,
         BalancerProtocols.class, fallbackToSimpleAuth).getProxy();
-    this.balancerShouldRequestStandby = conf.getBoolean(
+    this.requestToStandby = conf.getBoolean(
         DFSConfigKeys.DFS_HA_ALLOW_STALE_READ_KEY,
         DFSConfigKeys.DFS_HA_ALLOW_STALE_READ_DEFAULT);
-    this.standbyNameNode = null;
+    this.config = conf;
 
     this.fs = (DistributedFileSystem)FileSystem.get(nameNodeUri, conf);
 
@@ -216,24 +217,7 @@ public NameNodeConnector(String name, URI nameNodeUri, String nsId,
                            Configuration conf, int maxNotChangedIterations)
       throws IOException {
     this(name, nameNodeUri, idPath, targetPaths, conf, maxNotChangedIterations);
-    if (nsId != null && HAUtil.isHAEnabled(conf, nsId)) {
+    this.nsId = nsId;
-      List<ClientProtocol> namenodes =
-          HAUtil.getProxiesForAllNameNodesInNameservice(conf, nsId);
-      for (ClientProtocol proxy : namenodes) {
-        try {
-          if (proxy.getHAServiceState().equals(
-              HAServiceProtocol.HAServiceState.STANDBY)) {
-            this.standbyNameNode = NameNodeProxies.createNonHAProxy(
-                conf, RPC.getServerAddress(proxy), NamenodeProtocol.class,
-                UserGroupInformation.getCurrentUser(), false).getProxy();
-            break;
-          }
-        } catch (Exception e) {
-          //Ignore the exception while connecting to a namenode.
-          LOG.debug("Error while connecting to namenode", e);
-        }
-      }
-    }
   }
 
   public DistributedFileSystem getDistributedFileSystem() {
@@ -255,23 +239,43 @@ public BlocksWithLocations getBlocks(DatanodeInfo datanode, long size, long
     if (getBlocksRateLimiter != null) {
       getBlocksRateLimiter.acquire();
     }
-    boolean isRequestStandby = true;
+    boolean isRequestStandby = false;
+    NamenodeProtocol nnproxy = null;
     try {
-      if (balancerShouldRequestStandby && standbyNameNode != null) {
+      if (requestToStandby && nsId != null
-        return standbyNameNode.getBlocks(datanode, size, minBlockSize);
+          && HAUtil.isHAEnabled(config, nsId)) {
-      } else {
+        List<ClientProtocol> namenodes =
-        isRequestStandby = false;
+            HAUtil.getProxiesForAllNameNodesInNameservice(config, nsId);
+        for (ClientProtocol proxy : namenodes) {
+          try {
+            if (proxy.getHAServiceState().equals(
+                HAServiceProtocol.HAServiceState.STANDBY)) {
+              NamenodeProtocol sbn = NameNodeProxies.createNonHAProxy(
+                  config, RPC.getServerAddress(proxy), NamenodeProtocol.class,
+                  UserGroupInformation.getCurrentUser(), false).getProxy();
+              nnproxy = sbn;
+              isRequestStandby = true;
+              break;
             }
           } catch (Exception e) {
-      LOG.warn("Request #getBlocks to Standby NameNode but meet exception, " +
+            // Ignore the exception while connecting to a namenode.
-          "will fallback to normal way", e);
+            LOG.debug("Error while connecting to namenode", e);
-      isRequestStandby = false;
+          }
+        }
+        if (nnproxy == null) {
+          LOG.warn("Request #getBlocks to Standby NameNode but meet exception,"
+              + " will fallback to normal way.");
+          nnproxy = namenode;
+        }
+      } else {
+        nnproxy = namenode;
+      }
+      return nnproxy.getBlocks(datanode, size, minBlockSize);
     } finally {
       if (isRequestStandby) {
         LOG.info("Request #getBlocks to Standby NameNode success.");
       }
     }
-    return namenode.getBlocks(datanode, size, minBlockSize);
   }
 
   /**