Java Code Examples for org.apache.hadoop.hbase.zookeeper.ZKUtil#deleteNode()

private void unassignExcessMetaReplica(int numMetaReplicasConfigured) {
  final ZKWatcher zooKeeper = master.getZooKeeper();
  // unassign the unneeded replicas (for e.g., if the previous master was configured
  // with a replication of 3 and now it is 2, we need to unassign the 1 unneeded replica)
  try {
    List<String> metaReplicaZnodes = zooKeeper.getMetaReplicaNodes();
    for (String metaReplicaZnode : metaReplicaZnodes) {
      int replicaId = zooKeeper.getZNodePaths().getMetaReplicaIdFromZnode(metaReplicaZnode);
      if (replicaId >= numMetaReplicasConfigured) {
        RegionState r = MetaTableLocator.getMetaRegionState(zooKeeper, replicaId);
        LOG.info("Closing excess replica of meta region " + r.getRegion());
        // send a close and wait for a max of 30 seconds
        ServerManager.closeRegionSilentlyAndWait(master.getAsyncClusterConnection(),
            r.getServerName(), r.getRegion(), 30000);
        ZKUtil.deleteNode(zooKeeper, zooKeeper.getZNodePaths().getZNodeForReplica(replicaId));
      }
    }
  } catch (Exception ex) {
    // ignore the exception since we don't want the master to be wedged due to potential
    // issues in the cleanup of the extra regions. We can do that cleanup via hbck or manually
    LOG.warn("Ignoring exception " + ex);
  }
}
 

@Override
public void removeLastSequenceIds(String peerId) throws ReplicationException {
  String suffix = "-" + peerId;
  try {
    StringBuilder sb = new StringBuilder(regionsZNode);
    int regionsZNodeLength = regionsZNode.length();
    int levelOneLength = regionsZNodeLength + 3;
    int levelTwoLength = levelOneLength + 3;
    List<String> levelOneDirs = ZKUtil.listChildrenNoWatch(zookeeper, regionsZNode);
    // it is possible that levelOneDirs is null if we haven't write any last pushed sequence ids
    // yet, so we need an extra check here.
    if (CollectionUtils.isEmpty(levelOneDirs)) {
      return;
    }
    for (String levelOne : levelOneDirs) {
      sb.append(ZNodePaths.ZNODE_PATH_SEPARATOR).append(levelOne);
      for (String levelTwo : ZKUtil.listChildrenNoWatch(zookeeper, sb.toString())) {
        sb.append(ZNodePaths.ZNODE_PATH_SEPARATOR).append(levelTwo);
        for (String znode : ZKUtil.listChildrenNoWatch(zookeeper, sb.toString())) {
          if (znode.endsWith(suffix)) {
            sb.append(ZNodePaths.ZNODE_PATH_SEPARATOR).append(znode);
            ZKUtil.deleteNode(zookeeper, sb.toString());
            sb.setLength(levelTwoLength);
          }
        }
        sb.setLength(levelOneLength);
      }
      sb.setLength(regionsZNodeLength);
    }
  } catch (KeeperException e) {
    throw new ReplicationException("Failed to remove all last sequence ids, peerId=" + peerId, e);
  }
}
 

public void removeKeyFromZK(AuthenticationKey key) {
  String keyZNode = getKeyNode(key.getKeyId());
  try {
    ZKUtil.deleteNode(watcher, keyZNode);
  } catch (KeeperException.NoNodeException nne) {
    LOG.error("Non-existent znode "+keyZNode+" for key "+key.getKeyId(), nne);
  } catch (KeeperException ke) {
    LOG.error(HBaseMarkers.FATAL, "Failed removing znode "+keyZNode+" for key "+
        key.getKeyId(), ke);
    watcher.abort("Unhandled zookeeper error removing znode "+keyZNode+
        " for key "+key.getKeyId(), ke);
  }
}
 

private void cleanupHbckZnode() {
  try {
    if (zkw != null && hbckZodeCreated) {
      ZKUtil.deleteNode(zkw, hbckEphemeralNodePath);
      hbckZodeCreated = false;
    }
  } catch (KeeperException e) {
    // Ignore
    if (!e.code().equals(KeeperException.Code.NONODE)) {
      LOG.warn("Delete HBCK znode " + hbckEphemeralNodePath + " failed ", e);
    }
  }
}
 

@Test
public void testGetListOfRegionServers() throws Exception {
  // 0 region servers
  assertEquals(0, rt.getListOfRegionServers().size());

  // 1 region server
  ZKUtil.createWithParents(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname1.example.org:1234"));
  List<String> rss = rt.getListOfRegionServers();
  assertEquals(rss.toString(), 1, rss.size());

  // 2 region servers
  ZKUtil.createWithParents(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname2.example.org:1234"));
  rss = rt.getListOfRegionServers();
  assertEquals(rss.toString(), 2, rss.size());

  // 1 region server
  ZKUtil.deleteNode(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname2.example.org:1234"));
  rss = rt.getListOfRegionServers();
  assertEquals(1, rss.size());

  // 0 region server
  ZKUtil.deleteNode(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname1.example.org:1234"));
  rss = rt.getListOfRegionServers();
  assertEquals(rss.toString(), 0, rss.size());
}
 

@Test
public void testRegionServerRemovedEvent() throws Exception {
  ZKUtil.createAndWatch(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname2.example.org:1234"),
    HConstants.EMPTY_BYTE_ARRAY);
  rt.registerListener(new DummyReplicationListener());
  // delete one
  ZKUtil.deleteNode(zkw,
    ZNodePaths.joinZNode(zkw.getZNodePaths().rsZNode, "hostname2.example.org:1234"));
  // wait for event
  while (rsRemovedCount.get() < 1) {
    Thread.sleep(5);
  }
  assertEquals("hostname2.example.org:1234", rsRemovedData);
}
 

@Test public void testRestartMaster() throws IOException, KeeperException {
  ZKWatcher zk = new ZKWatcher(TEST_UTIL.getConfiguration(),
    "testActiveMasterManagerFromZK", null, true);
  try {
    ZKUtil.deleteNode(zk, zk.getZNodePaths().masterAddressZNode);
    ZKUtil.deleteNode(zk, zk.getZNodePaths().clusterStateZNode);
  } catch(KeeperException.NoNodeException nne) {}

  // Create the master node with a dummy address
  ServerName master = ServerName.valueOf("localhost", 1, System.currentTimeMillis());
  // Should not have a master yet
  DummyMaster dummyMaster = new DummyMaster(zk,master);
  ClusterStatusTracker clusterStatusTracker =
    dummyMaster.getClusterStatusTracker();
  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);
  clusterStatusTracker.setClusterUp();

  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);
  ActiveMasterManager secondActiveMasterManager =
    secondDummyMaster.getActiveMasterManager();
  assertFalse(secondActiveMasterManager.clusterHasActiveMaster.get());
  activeMasterManager.blockUntilBecomingActiveMaster(100, status);
  assertTrue(activeMasterManager.clusterHasActiveMaster.get());
  assertMaster(zk, master);
  assertMaster(zk, activeMasterManager.getActiveMasterServerName().get());
  assertMaster(zk, secondActiveMasterManager.getActiveMasterServerName().get());
}
 

private void unassignMetaReplica(HbckRegionInfo hi)
  throws IOException, InterruptedException, KeeperException {
  undeployRegions(hi);
  ZKUtil.deleteNode(zkw,
    zkw.getZNodePaths().getZNodeForReplica(hi.getMetaEntry().getRegionInfo().getReplicaId()));
}
 

private void deleteMyEphemeralNode() throws KeeperException {
  ZKUtil.deleteNode(this.zooKeeper, getMyEphemeralNodePath());
}
 

/**
 * This test will test the expire handling of a meta-carrying
 * region server.
 * After HBaseMiniCluster is up, we will delete the ephemeral
 * node of the meta-carrying region server, which will trigger
 * the expire of this region server on the master.
 * On the other hand, we will slow down the abort process on
 * the region server so that it is still up during the master SSH.
 * We will check that the master SSH is still successfully done.
 */
@Test
public void testExpireMetaRegionServer() throws Exception {
  MiniHBaseCluster cluster = TEST_UTIL.getHBaseCluster();
  HMaster master = cluster.getMaster();
  RegionStates regionStates = master.getAssignmentManager().getRegionStates();
  ServerName metaServerName =
    regionStates.getRegionServerOfRegion(RegionInfoBuilder.FIRST_META_REGIONINFO);
  if (master.getServerName().equals(metaServerName) || metaServerName == null ||
    !metaServerName.equals(cluster.getServerHoldingMeta())) {
    // Move meta off master
    metaServerName =
      cluster.getLiveRegionServerThreads().get(0).getRegionServer().getServerName();
    master.move(RegionInfoBuilder.FIRST_META_REGIONINFO.getEncodedNameAsBytes(),
      Bytes.toBytes(metaServerName.getServerName()));
    TEST_UTIL.waitUntilNoRegionsInTransition(60000);
    metaServerName =
      regionStates.getRegionServerOfRegion(RegionInfoBuilder.FIRST_META_REGIONINFO);
  }
  RegionState metaState = MetaTableLocator.getMetaRegionState(master.getZooKeeper());
  assertEquals("Wrong state for meta!", RegionState.State.OPEN, metaState.getState());
  assertNotEquals("Meta is on master!", metaServerName, master.getServerName());

  // Delete the ephemeral node of the meta-carrying region server.
  // This is trigger the expire of this region server on the master.
  String rsEphemeralNodePath =
      ZNodePaths.joinZNode(master.getZooKeeper().getZNodePaths().rsZNode,
              metaServerName.toString());
  ZKUtil.deleteNode(master.getZooKeeper(), rsEphemeralNodePath);
  LOG.info("Deleted the znode for the RegionServer hosting hbase:meta; waiting on SSH");
  // Wait for SSH to finish
  final ServerManager serverManager = master.getServerManager();
  final ServerName priorMetaServerName = metaServerName;
  TEST_UTIL.waitFor(120000, 200, new Waiter.Predicate<Exception>() {
    @Override
    public boolean evaluate() throws Exception {
      return !serverManager.isServerOnline(priorMetaServerName)
          && !serverManager.areDeadServersInProgress();
    }
  });
  LOG.info("Past wait on RIT");
  TEST_UTIL.waitUntilNoRegionsInTransition(60000);
  // Now, make sure meta is assigned
  assertTrue("Meta should be assigned",
    regionStates.isRegionOnline(RegionInfoBuilder.FIRST_META_REGIONINFO));
  // Now, make sure meta is registered in zk
  metaState = MetaTableLocator.getMetaRegionState(master.getZooKeeper());
  assertEquals("Meta should not be in transition", RegionState.State.OPEN, metaState.getState());
  assertEquals("Meta should be assigned", metaState.getServerName(),
    regionStates.getRegionServerOfRegion(RegionInfoBuilder.FIRST_META_REGIONINFO));
  assertNotEquals("Meta should be assigned on a different server", metaState.getServerName(),
    metaServerName);
}
 

/**
 * Unit tests that uses ZooKeeper but does not use the master-side methods
 * but rather acts directly on ZK.
 * @throws Exception
 */
@Test
public void testActiveMasterManagerFromZK() throws Exception {
  ZKWatcher zk = new ZKWatcher(TEST_UTIL.getConfiguration(),
    "testActiveMasterManagerFromZK", null, true);
  try {
    ZKUtil.deleteNode(zk, zk.getZNodePaths().masterAddressZNode);
    ZKUtil.deleteNode(zk, zk.getZNodePaths().clusterStateZNode);
  } catch(KeeperException.NoNodeException nne) {}

  // Create the master node with a dummy address
  ServerName firstMasterAddress =
      ServerName.valueOf("localhost", 1, System.currentTimeMillis());
  ServerName secondMasterAddress =
      ServerName.valueOf("localhost", 2, System.currentTimeMillis());

  // Should not have a master yet
  DummyMaster ms1 = new DummyMaster(zk,firstMasterAddress);
  ActiveMasterManager activeMasterManager =
    ms1.getActiveMasterManager();
  assertFalse(activeMasterManager.clusterHasActiveMaster.get());
  assertFalse(activeMasterManager.getActiveMasterServerName().isPresent());

  // First test becoming the active master uninterrupted
  ClusterStatusTracker clusterStatusTracker =
    ms1.getClusterStatusTracker();
  clusterStatusTracker.setClusterUp();
  activeMasterManager.blockUntilBecomingActiveMaster(100,
      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);
  t.start();
  // Wait for this guy to figure out there is another active master
  // Wait for 1 second at most
  int sleeps = 0;
  while(!t.manager.clusterHasActiveMaster.get() && sleeps < 100) {
    Thread.sleep(10);
    sleeps++;
  }

  // Both should see that there is an active master
  assertTrue(activeMasterManager.clusterHasActiveMaster.get());
  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");

  // Use a listener to capture when the node is actually deleted
  NodeDeletionListener listener = new NodeDeletionListener(zk,
          zk.getZNodePaths().masterAddressZNode);
  zk.registerListener(listener);

  LOG.info("Deleting master node");
  ZKUtil.deleteNode(zk, zk.getZNodePaths().masterAddressZNode);

  // Wait for the node to be deleted
  LOG.info("Waiting for active master manager to be notified");
  listener.waitForDeletion();
  LOG.info("Master node deleted");

  // Now we expect the secondary manager to have and be the active master
  // Wait for 1 second at most
  sleeps = 0;
  while(!t.isActiveMaster && sleeps < 100) {
    Thread.sleep(10);
    sleeps++;
  }
  LOG.debug("Slept " + sleeps + " times");

  assertTrue(t.manager.clusterHasActiveMaster.get());
  assertTrue(t.isActiveMaster);
  assertEquals(secondMasterAddress, t.manager.getActiveMasterServerName().get());

  LOG.info("Deleting master node");

  ZKUtil.deleteNode(zk, zk.getZNodePaths().masterAddressZNode);
}