HBASE-23275: Track active master's address in ActiveMasterManager (#812)

* HBASE-23275: Track active master's address in ActiveMasterManager

Currently we just track whether an active master exists.
It helps to also track the address of the active master in
all the masters to help serve the client RPC requests to
know which master is active.

Signed-off-by: Nick Dimiduk <ndimiduk@apache.org>
Signed-off-by: Andrew Purtell <apurtell@apache.org>

Author: bharathv

hbase-server/src/main/java/org/apache/hadoop/hbase/master/ActiveMasterManager.java

@@ -1,4 +1,4 @@
-/**
+/*
  *
  * Licensed to the Apache Software Foundation (ASF) under one
  * or more contributor license agreements.  See the NOTICE file
@@ -17,25 +17,24 @@
  * limitations under the License.
  */
 package org.apache.hadoop.hbase.master;
-
 import java.io.IOException;
+import java.util.Optional;
 import java.util.concurrent.atomic.AtomicBoolean;
-
-import org.apache.hadoop.hbase.zookeeper.MasterAddressTracker;
-import org.apache.hadoop.hbase.zookeeper.ZKUtil;
-import org.apache.hadoop.hbase.zookeeper.ZKWatcher;
-import org.apache.hadoop.hbase.zookeeper.ZNodePaths;
-import org.apache.yetus.audience.InterfaceAudience;
 import org.apache.hadoop.hbase.Server;
 import org.apache.hadoop.hbase.ServerName;
 import org.apache.hadoop.hbase.ZNodeClearer;
 import org.apache.hadoop.hbase.exceptions.DeserializationException;
 import org.apache.hadoop.hbase.monitoring.MonitoredTask;
-import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
+import org.apache.hadoop.hbase.zookeeper.MasterAddressTracker;
 import org.apache.hadoop.hbase.zookeeper.ZKListener;
+import org.apache.hadoop.hbase.zookeeper.ZKUtil;
+import org.apache.hadoop.hbase.zookeeper.ZKWatcher;
+import org.apache.hadoop.hbase.zookeeper.ZNodePaths;
+import org.apache.yetus.audience.InterfaceAudience;
 import org.apache.zookeeper.KeeperException;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
+import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
 
 /**
  * Handles everything on master-side related to master election.
@@ -57,12 +56,18 @@ public class ActiveMasterManager extends ZKListener {
   final AtomicBoolean clusterHasActiveMaster = new AtomicBoolean(false);
   final AtomicBoolean clusterShutDown = new AtomicBoolean(false);
 
+  // This server's information.
   private final ServerName sn;
   private int infoPort;
   private final Server master;
 
+  // Active master's server name. Invalidated anytime active master changes (based on ZK
+  // notifications) and lazily fetched on-demand.
+  // ServerName is immutable, so we don't need heavy synchronization around it.
+  private volatile ServerName activeMasterServerName;
+
   /**
-   * @param watcher
+   * @param watcher ZK watcher
    * @param sn ServerName
    * @param master In an instance of a Master.
    */
@@ -106,6 +111,30 @@ void handle(final String path) {
     }
   }
 
+  /**
+   * Fetches the active master's ServerName from zookeeper.
+   */
+  private void fetchAndSetActiveMasterServerName() {
+    LOG.debug("Attempting to fetch active master sn from zk");
+    try {
+      activeMasterServerName = MasterAddressTracker.getMasterAddress(watcher);
+    } catch (IOException | KeeperException e) {
+      // Log and ignore for now and re-fetch later if needed.
+      LOG.error("Error fetching active master information", e);
+    }
+  }
+
+  public Optional<ServerName> getActiveMasterServerName() {
+    if (!clusterHasActiveMaster.get()) {
+      return Optional.empty();
+    }
+    if (activeMasterServerName == null) {
+      fetchAndSetActiveMasterServerName();
+    }
+    // It could still be null, but return whatever we have.
+    return Optional.ofNullable(activeMasterServerName);
+  }
+
   /**
    * Handle a change in the master node.  Doesn't matter whether this was called
    * from a nodeCreated or nodeDeleted event because there are no guarantees
@@ -134,6 +163,9 @@ private void handleMasterNodeChange() {
           // Notify any thread waiting to become the active master
           clusterHasActiveMaster.notifyAll();
         }
+        // Reset the active master sn. Will be re-fetched later if needed.
+        // We don't want to make a synchronous RPC under a monitor.
+        activeMasterServerName = null;
       }
     } catch (KeeperException ke) {
       master.abort("Received an unexpected KeeperException, aborting", ke);
@@ -151,8 +183,8 @@ private void handleMasterNodeChange() {
    * @param checkInterval the interval to check if the master is stopped
    * @param startupStatus the monitor status to track the progress
    * @return True if no issue becoming active master else false if another
-   * master was running or if some other problem (zookeeper, stop flag has been
-   * set on this Master)
+   *   master was running or if some other problem (zookeeper, stop flag has been
+   *   set on this Master)
    */
   boolean blockUntilBecomingActiveMaster(
       int checkInterval, MonitoredTask startupStatus) {
@@ -178,10 +210,14 @@ boolean blockUntilBecomingActiveMaster(
           // We are the master, return
           startupStatus.setStatus("Successfully registered as active master.");
           this.clusterHasActiveMaster.set(true);
+          activeMasterServerName = sn;
           LOG.info("Registered as active master=" + this.sn);
           return true;
         }
 
+        // Invalidate the active master name so that subsequent requests do not get any stale
+        // master information. Will be re-fetched if needed.
+        activeMasterServerName = null;
         // There is another active master running elsewhere or this is a restart
         // and the master ephemeral node has not expired yet.
         this.clusterHasActiveMaster.set(true);
@@ -208,7 +244,8 @@ boolean blockUntilBecomingActiveMaster(
             ZKUtil.deleteNode(this.watcher, this.watcher.getZNodePaths().masterAddressZNode);
 
             // We may have failed to delete the znode at the previous step, but
-            //  we delete the file anyway: a second attempt to delete the znode is likely to fail again.
+            //  we delete the file anyway: a second attempt to delete the znode is likely to fail
+            //  again.
             ZNodeClearer.deleteMyEphemeralNodeOnDisk();
           } else {
             msg = "Another master is the active master, " + currentMaster +

hbase-server/src/main/java/org/apache/hadoop/hbase/master/HMaster.java

@@ -3829,6 +3829,10 @@ public String getClusterId() {
     return cachedClusterId.getFromCacheOrFetch();
   }
 
+  public Optional<ServerName> getActiveMaster() {
+    return activeMasterManager.getActiveMasterServerName();
+  }
+
   @Override
   public void runReplicationBarrierCleaner() {
     ReplicationBarrierCleaner rbc = this.replicationBarrierCleaner;

hbase-server/src/test/java/org/apache/hadoop/hbase/master/TestActiveMasterManager.java

@@ -17,6 +17,7 @@
  */
 package org.apache.hadoop.hbase.master;
 
+import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertFalse;
 import static org.junit.Assert.assertNotNull;
 import static org.junit.Assert.assertTrue;
@@ -91,6 +92,7 @@ public static void tearDownAfterClass() throws Exception {
     ActiveMasterManager activeMasterManager =
       dummyMaster.getActiveMasterManager();
     assertFalse(activeMasterManager.clusterHasActiveMaster.get());
+    assertFalse(activeMasterManager.getActiveMasterServerName().isPresent());
 
     // First test becoming the active master uninterrupted
     MonitoredTask status = Mockito.mock(MonitoredTask.class);
@@ -99,6 +101,7 @@ public static void tearDownAfterClass() throws Exception {
     activeMasterManager.blockUntilBecomingActiveMaster(100, status);
     assertTrue(activeMasterManager.clusterHasActiveMaster.get());
     assertMaster(zk, master);
+    assertMaster(zk, activeMasterManager.getActiveMasterServerName().get());
 
     // Now pretend master restart
     DummyMaster secondDummyMaster = new DummyMaster(zk,master);
@@ -108,6 +111,8 @@ public static void tearDownAfterClass() throws Exception {
     activeMasterManager.blockUntilBecomingActiveMaster(100, status);
     assertTrue(activeMasterManager.clusterHasActiveMaster.get());
     assertMaster(zk, master);
+    assertMaster(zk, activeMasterManager.getActiveMasterServerName().get());
+    assertMaster(zk, secondActiveMasterManager.getActiveMasterServerName().get());
   }
 
   /**
@@ -135,6 +140,7 @@ public void testActiveMasterManagerFromZK() throws Exception {
     ActiveMasterManager activeMasterManager =
       ms1.getActiveMasterManager();
     assertFalse(activeMasterManager.clusterHasActiveMaster.get());
+    assertFalse(activeMasterManager.getActiveMasterServerName().isPresent());
 
     // First test becoming the active master uninterrupted
     ClusterStatusTracker clusterStatusTracker =
@@ -144,6 +150,7 @@ public void testActiveMasterManagerFromZK() throws Exception {
         Mockito.mock(MonitoredTask.class));
     assertTrue(activeMasterManager.clusterHasActiveMaster.get());
     assertMaster(zk, firstMasterAddress);
+    assertMaster(zk, activeMasterManager.getActiveMasterServerName().get());
 
     // New manager will now try to become the active master in another thread
     WaitToBeMasterThread t = new WaitToBeMasterThread(zk, secondMasterAddress);
@@ -161,6 +168,8 @@ public void testActiveMasterManagerFromZK() throws Exception {
     assertTrue(t.manager.clusterHasActiveMaster.get());
     // But secondary one should not be the active master
     assertFalse(t.isActiveMaster);
+    // Verify the active master ServerName is populated in standby master.
+    assertEquals(firstMasterAddress, t.manager.getActiveMasterServerName().get());
 
     // Close the first server and delete it's master node
     ms1.stop("stopping first server");
@@ -189,6 +198,7 @@ public void testActiveMasterManagerFromZK() throws Exception {
 
     assertTrue(t.manager.clusterHasActiveMaster.get());
     assertTrue(t.isActiveMaster);
+    assertEquals(secondMasterAddress, t.manager.getActiveMasterServerName().get());
 
     LOG.info("Deleting master node");